This directory contains musical recordings (subdirectory "music"),
recorded research presentations ("talks") and recordings of many classes I\ve
taught at UCSD.  The classes are named by course number and date, so that
"206.16w" means "Music 206, 2016, winter quarter".  You can find corresponding
syllabi and other course materials, in, for instance,
msp.ucsd.edu/syllabi/206.16w .

Here are some notable ones:

170.21s  Musical Acoustics
171.20w  Computer Music
206.12f  Real-time computer music software design 
206.15w  Feedback and distortion
206.15f  Stored and live control streams for real-time electronic music 
206.18s  Frequency-domain audio techniques
206.19s  The voice as musical instrument 
206.20s  Inside Pure Data
270b.22s Analysis of musical sound
270c.21w Compositional Algorithms
271b.21s Survey of Electronic Music Techniques

COPYRIGHT NOTICE: unless otherwise noted, all media and supporting text files
are copyright Miller Puckette, available under the standard Creative Commons
share-alike license.  If you would like to share your own modifications (for
instance as part of other open courseware) please contact me and I\'ll be happy
to accomodate you.

What follows are tables of contents for some of the more recent videos, intended
for easy searching. The videos themselves may be watched online or downloaded.

-----------------------------------------------------
Music 170. Musical Acoustics.
Spring quarter 2021.
course materials: msp.ucsd.edu/syllabi/170.21s/
recordings: msp.ucsd.edu/media/170.21s/
contents of recordings:

msp.ucsd.edu/media/170.21s/170-01a-mar30.mp4: Meeting 1, March 30, 2021

0:00 syllabus - who should take this course and what you'll learn
5:30 about contract grading and how its used in this course
23:00 about the sound prompts (Ozzie and Harriet laugh and Wilhelm scream)
24:40 intro to Audacity sound editor, to play laugh track
29:11 Wilhelm scream sample
36:00 Pure Data and the acoustics library
37:00 spectrum of a sinusoid using Pd acoustics library
40:30 how to make patches using the Pd acoustics library
42:00 the path in Pd (how to get Pd to find the objects in the library)
43:00 the library as a directory
44:00 the file 1.LIST.pd - all the objects.  How to get help.
44:30 the "sinusoid" and "record" library objects.
45:00 edit mode and run mode in Pd
49:30 installing Pd on Microsoft windows
50:00 (three minutes of flailing)
54:00 downloading Pd on Windows
56:00 downloading the acoustics library
57:00 adding acoustics library to the Pd search path
58:00 oops, have to find where it went and unzip it
1:00:00 finally succeed in adding folder to path


msp.ucsd.edu/media/170.21s/170-01b-apr1.mp4: Meeting 2, April 1, 2021

0:00 incitement: dumb card trick
1;30 warning about in-ear headphones
3:30 first version of dumb card trick movie
4:30 how audio and image work together in a movie
5:00 version 2 of dumb card trick
6:00 version 3.
7:30 different functions of sound and image in a movie
10:00 more about the example and choices made in making it
14:00 why the smiley face?
17:30 more about video sound tracks
19:20 back to course organization and contract grading
24:00 make sure you have version 6 of Pd acoustics library
26:30 course "canvas" page
31:00 the Pd acoustics library: the "record" object again
32:30 setting length of the recording
34:00 listening to output of the "record" object
36:30 the "meter" object
37:00 using audacity to see the individual audio samples. Sample rate.
39:00 resolution (word size) of an audio recording
40:00 basic properties of a recording: sample rate, word length, number of channels
40:30 the PCM and compressed audio format (only use PCM for Pd or Audacity)
42:30 recording a sinusoid using the acoustics library
43:00 peak amplitude of a sinusoid.  Meaning of amplitude.
45:00 the meter shows average amplitude in two units (RMS and decibels)
45:30 frequency and starting phase (with amplitude, 3 properties of a sinusoid)
47:00 chapter 1 of the course notes: sinusoids
49:00 decibel scale
51:00 difference between acoustic amplitude and amplitude of a recording
52:00 two stages of conversion: sound/voltage and voltage/binary
54:00 what changing amplitude and frequency of a sinusoid sounds like
56:00 changing the amplitude and the frequency of a recorded waveform
57:30 the "multiply" object can change (scale) the amplitudes in a recording
1:01:00 mathematical description of changing frequency and amplitude
1:03:00 what "up" and "down" mean for amplitude and frequency
1:05:00 amplitude range of audio (-1 to 1 externally but freely variable in Pd)
1:06:00 average amplitudes shown by the "meter" object
1:07:00 multiplication by 0.1 decreases RMS by 0.1 and lowers decibels by 20
1:08:30 definition of RMS (root mean square) average
1:10:00 decibels are for describing proportions between amplitudes
1:12:00 more about decibels: multiple gain stages
1:13:30 what "gain" means
1:15:00 difference between dB (decibels) on meter and dB in "output" object
1:18:30 reminder: know your IA


msp.ucsd.edu/media/170.21s/170-02a-apr6.mp4: Meeting 3, April 6, 2021

0:00 incitement: phased looping inspired by Steve Reich's Piano Phase
3:00 reading sounds into "record" object: set length first
6:00 second copy in a second "record" object
7:00 playing the two loops together
9:00 relative phase of two sound loops
13:00 musical example: Charles Dodge, Any Resemblance Is Purely Coincidental
16:00 subject of the original opera aria ("Vesti la Giubba")
17:20 closer examination of the first 30 seconds of the piece
19:00 importance of the visual component of the piece
20:30 about the endlessly repeated word "recitar"
21:30 implications of using Caruso's voice in a 20th-century musical language
24:20 the dissected laugh
29:00 modern echos: "sampling"
31:30 ending: recognizable "money notes" and return (and fadeout) of orchestra
35:30 "great moments in opera" TV commercial
37:00 organization: back to the course canvas page; about projects and labs
51:00 More about using Pd: inputs and outputs and connections
52:00 why logarithms? demonstration of changing frequencies and amplitudes
59:30 sinusoid animation to see waveform, frequency and amplitude another way
1:02:00 starting phase (sine versus cosine)
1:03:00 arbitrary amplitude and starting phase
1:04:00 at a fixed frequency, sinusoids are all combinations of sin and cosine
1:06:00 raising the frequency by a factory of two
1:06:30 period of a waveform
1:08;00 adding in harmonics: sinusoids of multiples of original frequency
1:10:00 period of the resulting waveform
1:12:00 higher frequencies
1:13:00 setting frequency to halve the sample rate
1:14:00 foldover (frequency equal to sample rate is same as zero frequency)
1;17:00 maximum is half the sample rate. Standard rates are 44100 and 48000


msp.ucsd.edu/media/170.21s/170-02b-apr8.mp4: Meeting 4, April 8, 2021

0:00 review of contract requirements (A and B projects; all do 8 labs;
(C contract does one ninth lab instead of projects)
1:40 incitement: pulse width modulation made from sawtooth waveforms
5:00 spectrum of sawtooth wave
6:00 no perfect sawtooth wave in digital audio
7:30 you can hear a 55-Hz sawtooth wave even if your speakers can't play 55 Hz.
8:30 subtracting two waveforms using the acoustics library objects
10:30 rectangle waveform
11:00 single sawtooth versus difference between two of them
12:00 detuning the second sawtooth
14:00 spectrum of rectangle waves with changing pulse width
15:30 Two Women, Four Voices by Trevor Wishart and waveset repetition
17:30 first two minutes of piece (part 2, Princess Diana)
19:00 pitch correcting voice versus waveset repetition
20:00 recording from microphone to Audacity
21:00 changing the level of the recording
23:00 close-up of waveform of recorded voice
24:00 duplicating entire periods of vocal waveform
25:00 back to original recording, duplicating waveset insted of entire period
28:00 result of duplicating 3 successive wavesets
29:00 Audacity is very powerful - you can run any lisp function on the samples
30:00 stereo and mono soundfiles in Audacity
32:00 making a sinusoid in Audacity
34:00 (oops, mixed it down wrong)
34:00 select both tracks and reselect "mix and render to new track"
35:00 first mixing result
36:00 time-shifting one of the two sinusoids before mixing
37:00 third mixdown, another phase shift, lower amplitude output
39:00 beating between 1000-Hz sinusoid and 1002-Hz sinusoid
40:30 phase difference is changing in time between the two sinusoids
41:00 apparently changing the output's amplitude
41:30 "fundamental formula of computer music" (my private term for it)
43:30 applying the formula to the Audacity example to explain result
46:00 make beating effect by adding two sounds slightly out of tune
47:00 back to Pd, making a short vocal recording using acoustics library
48:00 applying gain to microphone input using "multiply" and "constant" objects
49:00 playing it back straight
50:00 multiply the voice recording by a sinusoid
51:00 sweeping frequency of sinusoid
52:00 try it with a sinusoid to see what happens to spectrum
54:00 This is called ring modulation
56:00 interference effect between two detuned sawtooth waves
1:00:00 For input sounds, you can use "vdelay" instead of detuning to change phase
1:02:00 doubling the perceived pitch by delaying by 1/2 period of a waveform
1:05:00 this is an example of a  filter.
1:07:00 white noise using the "noise" object in the acoustics library
1:08:00 filtering white noise using the "vdelay" object
1:09:00 objects used in the second lab assignment
1:09:45 graphical view of what happens when two sinusoids are added
1:11:00 different phases give different gains
1:13:00 detuning one sinusoid makes the phase difference change in time
1:13:30 120 degree (1/3 cycle) phase difference, same amplitude out as in
1:14:00 what this implies for incoming noise signal
1:16:00 analogy with light: diffraction gratings and oil films on puddles
1:17:30 sound engineers call this a comb filter


msp.ucsd.edu/media/170.21s/170-03a-apr13.mp4: Meeting 5, April 13, 2021

0:00 incitement: paulstretch "effect" in audacity applied to Wilhelm scream
2:00 different quality of sound when stretched out
2:30 pitch comes through but "presence" is lost
4:00 how to fade sounds in and out
4:30 fading is done using multiplication (applying time-verying gain)
5:00 paulstretch on ozzie-harriet laugh track
7:00 cackles at end of laugh track
7:30 cross-fade between the two stretched sounds
9:00 hmm, timing of crossfade is too abrupt
10:00 changing timing of crossfade (alignment and fade times)
11:00 artistic example: Jacqueline George, Just a Market
12:45 historical context
13:30 how to make political statements without being thrown in jail
15:30 coughing sound emerges from crowd
16:30 function of repetition effect in the piece (presaging a later event)
17:30 ending of piece, automatic rifle fire echoes cough
19:00 keeping meanings ambiguous
20:00 genre is soundscape composition
21:30 more specifically narrative soundscape composition (Yvette Jackson's term)
23:00 preparation in the piece is a kind of repetition
24:00 what happens when a sound comes back in a piece later in time
25:30 what to start with (be clear on purpose, or maybe do the exact opposite)
27:30 the opposite approach: Eno and Schmidt's Oblique Strategies
31:00 theory and technique: more on filtering
31:20 preparatory aside: recording level and gain stages.
31:40 saturation (clipping)
32:30 dynamic range of a signal
33:00 bottom of range is (usually) noise floor
33:30 difference between normal speech level and background noise
34:00 sound being recorded should be well above noise floor
35:00 top of dynamic range (loudest signal achievable)
35:45 signal to noise ratio, in dB.  Dynamic range of human hearing is 100dB
37:00 stages of conversion, each (ideally) with a variable input gain
38:00 even though we're now streaming, it's also a form of recording
39:00 clipping at input of interface and its noise floor
39:30 aligning dynamic range windows at two stages of conversion
41:00 find the stage with the smallest dynamic range
42:30 easy way to line these up is to measure or set headroom
44:30 if you clip or get too much noise there's no easy way to fix later
44:45 not a good strategy to go looking through things on the web
46:00 microphone placement
47:30 getting the mic close to the source improves dynamic range of recording
49:00 try this experiment with your phone
50:00 back to incitement: time stretching using the acoustics library
52:30 this is an impoverished time stretcher, but OTOH it's real time
54:00 filters in the acoustics library
55:00 the low-pass filter ("lowpass" object)
56:00 spectrum of low-pass filtered noise
57:30 frequency response (gain as a function of frequency) of a low-pass filter
58:00 0 dB gain at low frequencies, and a settable rolloff frequency
59:00 band-pass or resonant filter: center or resonant frequency
59:30 gain at center frequency (peak gain), not always 0dB in practice
59:45 a 0 dB gain is a "unit gain" (0 dB is unity).  Linear ("RMS") gain is 1.
1:01:00 bandwidth of a band-pass filter
1:02:00 "Q" ("quality") of a bandpass filter.  Q of zero means no filtering
1:02:40 Q is defined as (center frequency) divided by bandwidth
1:03:00 why we use "Q" as a measure of filter bandwidth
1:04:00 band-pass filter with center frequency 1000 Hz, Q=10 (bandwidth 100 Hz.)
1:05:00 watch out when you decrease Q with speakers open
1:05:30 set level with Q=0, then increase Q if you want.
1:06:00 what Q=100 sounds like
1:06:45 filtering a sinusoid
1:08:00 using "meter" object to measure output level as function of frequency
1:08:45 band-pass filter applied to sawtooth at 2 Hz.
1:09:00 what a 2-Hz. sawtooth sounds like
1:09:30 sawtooth at 2 Hz through a bandpass filter with  Q = 100 or 1000
1:11:00 recording of bandpass output.  Exponentially decaying sinusoid
1:13:00 Ozzie-Harriett-laugh through band-pass filter, Q=10
1:15:00 filtering output of paulstretched laughter
1:16:00 making a major chord out of it
1:17:00 band-pass filters can impose pitches on recorded sounds


msp.ucsd.edu/media/170.21s/170-03b-apr15.mp4: Meeting 6, April 15, 2021

0:00 incitement: feedback as oscillator
1:45 need an object to clip the sgnal (control runaway amplitudes)
3:00 any real object will saturate (clip) at some input level 
3:30 clipping changes teh shape of a waveform
4:30 delay in the feedback path
5:00 need a filter in the feedback path
5:30 feedback gain greater than one causes unstable feedback
6:00 "between" object does the clipping
7:45 talking into the feedback patch
8:30 center (peak) frequency of bandpass filter affects pitch of the output
9:00 higher value of Q makes the system more selective
9:30 tuning fork to inject sinusoid into the system
10:00 tuning fork fights with circuit over which frequency to resonate at
11:00 adding extra delay (to whatever delay the system already has)
12:30 example of this in popular music: 
13:00 (you heard the feedback example through the mic only)
13:30 Forbidden Planet soundtrack example
15:45 soundtrack is both music and sound effects (usually thought of separately)
16:20 movie soundtrack composers have more freedom than pop musicians
18:00 feedback patch makes similar sounds but the technology was different
19:00 example: guitar intro to "it's all too much" by the Beatles
20:30 filtering by guitar strings and electronics; delay was through the air
22:30 Jimi Hendrix at Woodstock
23:30 Jimi reaches for the wahwah pedal (resonant / bandpass filter)
24:00 clippnig is in vacuum tubes in amplifier; delay is physical distance
25:45 software oscillators work usnig feedback in the code
26:30 analog oscillators also work via feedback
26:45 analog synthesis example: Wendy Carlos, Switched-on Bach
29:00 voltage-controlled synthesizer controlled by clavier (music) keyboard
30:00 oddness of idea of maknig a popular record album out of very old music
32:30 same techniques are available digitally
33:00 basic sound source is sawtooth wave and its relatives
34:30 filtering a sawtooth wave
35:30 aside: typing into mumber boxes in Pd
36:00 filtering the sawtooth wave usnig "bandpass" object
37:00 sweeping the filter with Q=10
38:00 lower values of Q
39:00 interesting idea: controlling resonant frequency with an audio signal
40:00 using a multiplier to scale output of the controlling sawtooth
41:00 you can turn the sound off by setting filter center freqency very low
41:30 displaying output of controlling sawtooth generator
43:00 making repeated notes by filtering
44:30 multiple sawtooth generators to make richer sound
45:30 octave is a ratio of 2 to 1 in frequency
45:45 musical interals are set by ratios between frequency
47:00 sawtooth generators in the ratio 1:2:3:4:5 in frequency
47:30 3:2, 4:3, 5:4 ratios as intervals
48:30 shift key while making connection to make multiple connections at once
49:00 the five together sound ambiguously like one or several tones
49:30 spectrum of teh sum
50:30 spectrum of two sawtooth waves in ratio 5:2
51:00 when do you hear the combination as a single pitch?
52:00 detuning teh 5:2 ratio
53:00 fact that 5th, 10th, etc partials are changing draws attention to them
54:00 fixed spectrum graph to agree with audible output
54:30 detuned combinations of waveforms, popular technique in analog synths
55:30 detuned sum into time-varying resonant filter
56:00 clavier keyboard controlling filter using transient (envelope) generator
57:00 another voltage output controls frequency of oscillators
58:30 this is called subtractive synthesis
59:00 music 171 will take this further
1:00:00 acoustics library doesn't do feedback (for the sake of simplicity)
1:01:00 detunig frequencies occur automatically if we use Western musical scale
1:02:50 terminology: harmonic and inharmonic spectra
1:03:30 usnig a filter with very high Q to isolate individual sinusoids
1:04:30 discrete versus continuous (noisy) spectra
1:06:00 harmonic versus inharmonic discrete spectra
1:07:00 spectrum of a bell sound
1:08:00 unevenly spaced components (sinusoidal frequencies in spectrum)
1:09:00 using a filter to examine spectrum of the bell
1:10:30 heard pitch is that of a 141-Hz. sawtooth (this is at the wrong speed)
1:11:00 392 Hz. found frequency is not a multiple of 141.
1:11:30 171 Hz. about a minor third above heard pitch
1:13:00 there might be beating pairs of components in bell
1:16:00 open question, how to make spectra that sound like well-designed bells
1:18:00 idea: use a sinusoid to change the frequency of a sawtooth
1:19:00 even-numbered components are moving in parallel
1:19:30 your ear then hears two sawtooth waves as separate sounds
1:22:00 components can be close to each other and beat
1:23:00 most components of circular bells are in pairs that are slightly detuned


msp.ucsd.edu/media/170.21s/170-04a-apr20.mp4: Meeting 7, April 20, 2021

0:00 incitement: real-time time reversal using the acoustics library
1:00 variable delay time makes Doppler shift
2:00 vibrato from time-varying vdelay using sinusoid to control delay time
3:00 different from vibrato by controlling the oscillator
4:00 variable-delay treatment of incoming sound
7:30 very deep pitch shifts, even making frequency go negative
8:45 diagram shoing how variable delays change frequency
10:00 extreme case where signal flips around backward in time
12:30 voice into extreme vibrato so that it reverses in time
13:00 using a sawtooth wave (instead of sinusoid) to control delay time
13:30 shifting downward only
14:30 real-time reversal of incoming signal
17:00 example: radiohead, everything in its right place
20:30 example: Hendrix guitar solo from Purple Haze - speed doubling
23:30 manipulating expectation in time-based art
26:30 prolongation: doing something later than expected
27:00 prolongation in pop lyrics
28:00 example: Arlo Guthrie, Pickle
32:00 example: Billie Eilish
35:00 another thing about Eilish phrase: loudness balance
36:00 balancing amplitude didn't balance loudness at all
36:45 10-ish decibels is not a 10x difference in loudness
37:00 equal loudness contours
38:00 axes in plot: dB versus frequency (both log scales)
40:00 how loud does a 100-Hz tone have to be to balance 80 dB at 1000 Hz?
41:00 hearing and voice seem to have co-evolved
43:00 threshold of hearing
43:30 loudness differences more perceptible at low frequencies
45:00 need much more sound pressure at low frequencies for same loudness
46:00 spectrum of Eilish voice: peak around 300 Hz.
49:00 tone right after vocal phrase, 50 Hz.
51:00 need 15-ish dB more at 50 Hz. to match 300 Hz.
52:00 the 50-Hz. tone was as strong as apparently permitted
54:00 clarification from last time: harmonics, overtones, partials, components
55:30 periodic waveforms
1:00:00 why we care about theories of loudness
1:00:30 non-definition of "timbre" of a sound
1:02:00 how ears work (very crude caricature)
1:02:30 the cochlea
1:03:00 basilar membrane and nerve cells that fire in time with its motion
1:04:30 why the filtering done by the basilar membrane is important
1:05:30 sounds of different frequencies travel different distances down it
1:07:00 the lower the frequency, the further it gets
1:08:00 typical spatial curves of excitation along basilar membrane
1:09:00 excitation at fixed location as function of frequency
1:10:00 different locations get different frequency responses
1:10:45 graph: Slaney Audio Toolkit estimate of basilar membrane excitation
1:12:00 implication: at a fixed frequency, there's a rapid drop-off
1:13:00 slightly lower frequencies aren't rejected as well as slightly higher ones
1:14:00 critical bands, bandwidth of frequency response at a point on membrane
1:15:00 within a critical band stuff adds in units of power
1:15:30 more widely separated frequencies contribute more to loudness
1:17:00 pitch acuity for flat versus sharp notes, and soft versus loud tones


msp.ucsd.edu/media/170.21s/170-04b-apr22.mp4: Meeting 8, April 22, 2021

0:00 incitement: sample and hold
2:00 noise as input to sample and hold to control oscillator frequency
5:30 decimation (one form of bit crushing)
8:00 decimating a voice recording
13:00 reasons for using simple ("low tech") techniques
14:00 artistic example: another kind of prolongation from Hitchcock, The Birds
16:30 soundtrack electronically composed (not ordinary bird recordings)
19:00 foreground and background sounds (sound changes when we enter school room)
20:00 children's song builds suspense by repetition
21;00 gathering crows effect reinforced by soundtrack
23:00 how the brain buids connections between senses: the McGurk effect
25:00 you are being made to hear the same sound differently because of image.
27:30 more on psycoacoustics: accuracy of pitch perception
29:30 basilar membrane excitation from an incoming sinusoid
30:00 the just noticeable difference in pitch
31:00 two sinusoids at 440 and 441, inaudible difference
31:30 440, 445 (1 percent) difference was decently audible
33:30 there are ten octaves in the range of hearing
34:30 in ideal conditions we can distinguish about 2000 different pitch values
35:00 how can you hear frequency so accurately given basilar membrane filters?
36:00 one possible answer: right-side edge of excitation curve drops sharply
37:00 the place theory of pitch perception
38:00 the alternate explanation: timing of nerve firings
41:00 pitch perception is one of the miraculous things about human hearing
41:30 another miraculous thing: the dynamic range of hearing
43:00 10 billion to one ratio of power between loudest and softest sound
43:30 another miraculous thing: the coctail party effect
45:30 the gestalt theory of perception applied to hearing
46:30 how the brain makes percepts
47:30 some grouping cues: common features; common fate
49:00 common-fate Gestalt principle in hearing: how partials can fuse into one sound
49:30 applying vibrato to even-numbered partials (idea is from Stephen MacAdams)
51:00 graphing the spectrum
54:30 sound of the odd harminics alone
55:00 you might hear the timbre of the lower tone chain change
56:00 applying vibrato to the odd harmonics only instead of the even ones
57:30 perception of loudness
1:00:00 6 decibels is roughly one notch louder to musicians
1:00:30 musicians use the word "dynamic" to mean "force"
1:01:00 the musical symbols for dynamics (ppp, pp, p, mf. f. ff, fff)
1:02:00 if we consider one "dynamic" as 6 dB, then 36 dB range would suffice
1:03:00 36 dB is 64:1 in RMS amplitude or 4096:1 in power
1:04:00 definition of a sone: experimentally, how loud is "twice as loud"?
1:05:30 the word "level" is code for decibels.
1:06:00 loudness "doubles" if you raise it by 10 decibels
1:06:00 relative loudness in sones is 2 ^ ((L - L1)/10)
1:08:00 ISO definition of loudness in sones: [relative RMS amplitude] ^ 0.3
0:09:00 consider threshold of hearing at 1000 Hz. as a reference loudness
1:09:30 loudness of an arbitrary sinusoid in sones
1:10:30 answer for this example: 8 sones
1:11:00 1000 sones in hearing range (because 100 dB gives 10 doublings)
1:12:00 some uses for sones
1:13:00 how VU meters and sliders are marked off in audio equipment
1:18:30 what does doubling loudness really mean?


msp.ucsd.edu/media/170.21s/170-05a-apr27.mp4: Meeting 9, April 27, 2021

0:00 syllabus change: adding a week on "voice"
2:00 2-part incitement: draw your waveform, and dueling metronomes
4:30 second metronome has period golden ratio times that of the first
6:30 two cycles superimposed so you can't tell which is which
7:30 this is a way to be neither random nor predictable
9:30 this works at different time scales and on the same or different layers
12:00 the music of Harry Partch using special, just-intoned instruments
17:15 Charles Corey describes one of the instruments, the harmonic cannon
20:00 consonant and dissonant intervals
21:00 ranges of intervals: beating; roughness; intervals
22:30 spectrum of sum of two complex harmonic tones
23:30 integer-ratio intervals sound consonant
25:00 partials of combination of tones with rational ratio
26:00 layout of Partch diamond marimba
27:00 six blades on SW/NE diagonals are in ratio of 4:5:6:7:9:11
28:00 NW/SE blades are in ration 1/4:1/5:1/6:1/7:1/9:1/11
30:00 building 4:5:6:7:9:11 scale in a patch
34:00 NW/SE scale
37:00 Partch scale (diamond marimba plus some extra tones; 43 in all)
39:30 what's special about integer ratios
41:00 spectrum of two sinusoids added
42:00 critical band theory of dissonance (roughness) for two sinusoids
43:00 critical bands are about 20% wide above 500 Hz; 100 Hz below it
46:30 discordant combination of 1000, 1050 Hz sinusoids
47:30 smaller than 30-ish Hz gives beating; nastiness kicks in around 10
48:30 difference tones
49:00 setting intervals to avoid roughness among partials
49:30 close together or far apart is OK, but in between sounds "rough"
50:00 200 Hz. versus other tones
51:00 terms: consonance and dissonance
52:00 consonance/dissonance of different ratios
54:30 Helmholz theory of consonance/dissonance
58:00 perfect versus almost-perfect intervals
59:30 1/2 way between 6:5 and 5:4 ratios
1:02:00 building out a scale using consonant intervals
1:03:00 3:2 ratio
1:04:30 music theory and the major triad
1:05:00 triad is 3-note chord with base pitch, plus major third and fifth
1:05:30 names of intervals: 6/5 minor third, 5:4 major third; 3:2 fifth
1:06:00 sound of major triad
1:07:00 how to build a Western musical scale out of triads
1:08:30 a fifth is combination of a major and a minor triad
1:10:00 adding the third triad to give a 7-tone scale
1:11:30 a sour interval in the scale (almost but not a fourth
1:12:00 fourth is 4:3 ratio
1:12:30 sour intervals are particularly problematic in polyphonic music
1:15:00 either the fourth is bad or the fifth is.
1:16:00 splitting the difference to reduce the dissonance
1:17:00 Western major scale is made out of triads with adjustments
1:19:00 scale comes from some combination of innate hearing and culture


msp.ucsd.edu/media/170.21s/170-05b-apr29.mp4: Meeting 10, April 29, 2021

0:00 incitement: feedback delay networks ("FDNs")
9:00 mixing delay outputs back to their inputs
13:00 non-repeating echo pattern
15:30 musical example: ending of Beethoven Hammerklaviersonate - cadences
19:00 dominant and tonic chords
23:00 cadences aren't just playing same chord in parallel (some tones go up)
25:00 the whole sequences of not-quite-cadences
28:00 notes on which "false cadences" land aren't climbing by same interval
30:00 the piece can't just stay on the tonic chord - you need to delay resolution
32:00 music works in the context of expectations raised by previous music 
34:30 what do you get when you reverse that last cadence?
35:30 pulse generator (was used in lab but I forgot to introduce it in class)
38:00 spectrum controlled by one parameter names bandwodth
39:00 pulses in time domain (as graphed in "record" are related to bandwodth
42:00 correction from last time: perfect interval between 5:$ and 6:5 (11:9)
46:00 musical scales
47:00 stacking four perfect fifths on top of each other: 81/64 interval
48:30 9:8 interval: major second
50:00 how intervals are named: unison, second, third etc.
51:30 two "seconds" don't make a perfect third
52:45 it's off by 1.2 percent in frequency, more than just noticeable difference
53:30 the interval from 5/4 to 81/64 (syntonic comma, although I didn't say it)
54:30 practical way of resolving the mis-tuning
55:00 the fourth: 4/3 ratio
56:00 interval from third to fourth note is only about 6% (a second was 12.5%)
57:30 the rest of the scale (fifth onward)
58:00 12% is twice as wide as 6%, so why not stick extra notes between seconds
59:00 why the piano keyboard looks the way it does
1:00:00 punch line: if you want to have a scale starting from anywhere...
1:01:30 (half-step is name of the smaller interval)
1:02:30 western major scale is (2, 2, 1, 2, 2, 2, 1) half-steps
1:03:00 how pitches are named
1:04:00 set a number "h" to the ideal half-step ratio.  h^12 = 2
1:05:00 "h" is 1/12 if an octave, so twelfth root of two.
1:06:00 how many half-steps go in a perfect third?  "fixed" third is 0.8% high
1:07:00 the fifth is 0.1% off
1:08:00 how many octaves in a perfect fifth?  log(3/2)/log(2), about 7/12
1:13:00 perfect third is 14% shy of 4 half steps
1:14:00 number of half-steps in the 81/64 "third" - 8% too much
1:15:30 These h-based intervals are called "tempered"
1:16:00 how the three thirds sound
1:16:00 why are there 12 notes in the Western octave?  Two possible answers
1:18:00 12-tone tempered scale has pretty good approximations of 3/2 and 5/4
1:20:00 idea: divide octave into different numbers of equally spaced pitches


msp.ucsd.edu/media/170.21s/170-06a-may4.mp4: Meeting 11, May 4, 2021

0:00 incitement and artistic example: Lucier, I Am Sitting in a Room
13:30 modes of vibration (in an open-ended tube or in a room)
19:30 process music (a particular spin on minimalizm)
23:30 how the use of space in music is impacted by physical isolation
27:30 what Zoom does to sound
30:30 working toward question of what happens when sound bounces off a wall
32:00 sound bounces off barriers (and off open ends of tubes in a different way)
33:00 microphones don't pick up spatial distribution of sounds
34:00 understanding how sound travels helps you plan recording and playback
36:00 sound pressure variations over space
37:00 sinusoidal plane waves, in space and seen along the direction of travel
39:00 the velocity of air motion in a plane wave
40:30 pressure as picked up at a fixed point in space - function of time
42:00 period of sinusoidal signal related to frequency
43:00 wavelength (lambda)
47:00 formulas relating wavelength, frequency, and period
48:30 speed of sound ~ 1100 feet per second; 1000 Hz. gives wavelength of 1 foot
52:00 time of travel of sound through air
53:30 measures of sound strength - intensity and sound pressure level
53:00 sound radiation in terms of pressure and velocity (scalar and vector)
56:00 power and intensity
57:00 possible units of intensity are watts per square meter
57:30 definition of flux
58:00 intensity (flux across an area) is a vector
59:00 conservation of power implies intensity drops with distance r as (1/r^2)
1:01:00 intensity related to RMS sound pressure: I = p^2 / (rho c)
1:03:00 how wattage relates to sound pressue level
1:04:00 nominal sound pressure level that is considered 0 dB
1:05:00 those equal-loudnes contours are normalized to this level
1:07:00 your ear can't withstand much pressure variation at audible ferquencies
1:08:00 intensity of a plane wave is directed in the direction of the wave
1:09:00 a full description of the sound field at a point would include all directions
1:11:00 how the 3 velocity components depend on complete sound field
1:13:00 measurable sound at a point (ignoring its neoghbors) is only 4-dimensional
1:14:30 implications for recording


msp.ucsd.edu/media/170.21s/170-06b-may6.mp4: Meeting 12, May 6, 2021

0:00 incitement: automatic beat boxer using the bonk~ object
7:00 also using a vocoder (will look at that next week)
9:00 sub-incitement: you can make patches into VST plug-ins
9:30 artistic example: Janet Cardiff, sound walks
23:00 back to plane waves and directionality: plane waves bouncing off walls
23:00 planes of maximum pressure at an instant in time in a sinusoidal plane wave
24:30 what the pressure looks like as a function of time
25:00 sound has no polarization (unlike light)
27:00 aside: how sound travels through the earth (solid, except its core)
30:30 incitement (PhD level) make the "earth" plug-in 
33:00 (back to sound in the air) bouncing a plane wave off an infinte planar barrier
36:00 planes where pressure is not changing and others where air is not moving
40:00 nodes are places where there is no pressure variation
43:00 standing waves in an air column
47:00 plane waves reflecting at an angle
48:00 how to use this to measure combinations of velocity and pressure
48:30 omnidirectional microphones only measure pressure, no velocity
49:00 cardioid microphone design using an omni mic and a small reflector
50:00 (back to plane wave reflection) peak pressure change is at the wall
52:30 cardioid equation r = cos(theta)+1
54:00 demonstration: omni microphone near a reflector
56:00 distance from a mic alters balance between direct and reflected sound
57:00 room acoustics in the modal versus the geometric view
58:00 modes in a room are lower-frequency and more numerous than in a flute
59:00 geometric view focuses on reflection
59:30 modes and nodes in the modal view
1:01:00 directional versus omnidirectional mics
1:02:00 different balances between direct and reflected sounds
1:03:00 recording an instrument (mic placement)
1:05:00 far-field guitar recording using an omni mic
1:05:30 proximity effect (but also distorted; that wasn't intentional)
1:07:00 how a guitar radiates sound idealized as a moving rigid plate
1:08:00 low-frequency sound diffracts around from the back of the plate
1:09:00 so low frequencies are attenuated by cancelation
1:09:30 at higher frequencies there is less attenuation
1:10:00 but if you're too close the low-frequency attenuation is reduced.
1:10:30 the far-away sound is considered the "true" sound of the instrument
1:11:00 the close-up sound emphasizes low frequencies compared to far field
1:12:00 problem in recording: get mic away from instrument but keep a clear sound
1:15:00 emitted sound is more directional at high frequencies than at lows
1:15:30 physics of diffraction
1:17:00 angle at which far-field emitted sound drops off (L sin theta = lambda)
1:18:00 in other words, theta (beam width) is about lambda/L in radians
1:19:30 light wavelengths are much smaller so we see much less diffraction
1:20:00 at 20 Hz you'd need a 500-foot wide emitter for a beam 1/10 radian wide


msp.ucsd.edu/media/170.21s/170-07a-may11.mp4: Meeting 13, May 11, 2021

0:00 incitement: vocoders ("channel" vocoders, not phase vocoders)
5:30 linearity of filter input to vocoder (can superpose two inputs)
6:30 vocal production mechanism: glottis and resonant cavities
7:30 glottis can be thought of as a pulse generator
9:30 fricative consonants result from turbulence, and don't involve glottis
11:00 vocoder models the vocal tract filter
12:00 "whispering" by sending noise into vocoder
13:30 example: Laurie Anderson's "O Superman"
17:00 the "Bell Labs" vocal model and what you can do with it
19:00 modeling the vocal resonator as a tube stopped at one end
20:00 possible standing waves in an air column 
21:00 what happens when the air column is closed at one end
23:00 testing a metal tube as an air column 
25:00 6 inch open tube makes peaks at 1000, 2000, 3000 Hz
26:00 spectrum of 1/2-stopped tube: 500, 1500, 2500, ... Hz.
27:00 open tube wavelengths: 2L, 2L/2, 2L/3, ...
28:00 half-closed tube: 4L, 4L/3, 4L/5, etc.
30:00 a real resonator in enclosed space (hard to hear the effect)
32:30 same thing, using reciprocity (placing mic in the enclosure)
35:00 spectrum of shaker lowered into enclosure (more audible this time)
36:00 spectrum of vocal sounds: resonances
37:00 neutral vowel resonances around 500, 1500, 2500 Hz.
38:00 in human speech, resonator constantly changes shape
39:00 up/down and forward/back vowels
40:00 schwa (neutral vowel) is in middle
41:30 spectral peaks at sinusoidal components
42:00 formants, peaks in the heights of the sinusoidal peaks
43:00 reinforced harmonics in singing
45:00 formants don't necessarily lie on harmonic frequencies
46:00 formants determine the vowel (to a first approximation)
46:30 lowest formant approximates front-back position of tongue
47:00 this can be better analyzed as a Helmholz resonator
48:00 low-frequency peak isn't considered a formant
50:00 higher-frequency formants are less clearly resolved
51:30 problem forming vowels when singing high pitches
56:30 ways real vocal production don't follow the simple model
59:00 high harmonics are less well behaved than low ones
60:00 glottal pulses shaped differently from each other
1:01:00 aperiodicity of glottal pulse train mostly at high frequencies
1:02:30 fricative consonants primarily high freqency
1:05:30 pitch of singing versus speaking not as different as you might think
1:07:30 psychoacoustic effect of repeating a spoken uterance (Diana Deutsch)
1:11:00 whether something is understood as singing depends in context
1:15:00 conundrum: how does singing work in a tonal language?


msp.ucsd.edu/media/170.21s/170-07b-may13.mp4: Meeting 14, May 13, 2021

0:00 incitement: phase vocoder, vocal recording up close
3:00 frozen sound (infinite time stretching)
4:30 how the phase vocoder works: splitting a sound into sinusoids
5:00 abusing the phase vocoder with an input that doesn't obey the model
5:30 the window size controls the available ferquency resolution
6:00 1/10 second window, 4096 oscillators playing the soind back
7:00 how it sounds with a much smaller window
9:00 trade-off, hazy sound from a larger window
9:30 what a shwa vowel soundls like with a very small window
10:45 musical example: voice morphs into swarming bees (Trevor Wishart, Vox #5)
14:30 this used the phase vocoder, part of the Composer's desktop project
16:30 radio announcer up close as a waveform
17:30 near-periodicity during voiced portions (vowels)
20:00 non-periodic "s" consonant
22:30 imitating voiced sounds using pulse generator
24:30 time-domain pulse signal (i.e. pulse train as function of time)
25:00 applying bandpass (resonant) filters
29:00 close-up recording of a single pulse through the filter
31:30 increasing q from 10 to 100 so taht you hear resonance as a pitch
33:30 higher value of q makes resonance (ringing)  last longer
35:00 filter frequency response versus waveform of output (amplitude vs. time)
37:00 why people use "q" instead of just specifying bandwodth
38:30 decay time of response
39:30 decay time is inversely proportional to bandwodth
40:00 so q is (loosely speaking) the number of cycles the resonance lasts
43:00 spectrum and time response for q=1, 10, 100, and 1000
46:00 decaying resonance in the vocal recording in Audacity
49:00 appear to have resonances at 700 and 4000 Hz.
49:30 verifying this using Audacity's spectrum display
52:00 attempt to reconstruct a vowel using pulse and filters
55:00 whether to put filters in series or in parallel
1:01:00 seat-of-the-pants alternative: filters in parallel
1:03:00 vowel shape maintained even if changing the pitch (pulse frequency)
1:03:30 replacing pulse train with noise source
1:06:00 real vocal sounds noisier at higher frequencies
1:07:30 changing the vowel sound - vowel is characterized by formant frequencies
1:12:00 speech (production and understanding) has many practical applications
1:13:00 glottis + air column picture revisited
1:15:00 standing waves inside air column correspond to resonances
1:16:00 each resonance should die out in time between glottal pulses
1:16:30 ... unless relaxation time is bigger than time between pulses


msp.ucsd.edu/media/170.21s/170-08a-may18.mp4: Meeting 15, May 18, 2021

0:00 incitement: Shepard tone
7:00 foldover
8:00 rising instead of falling
10:00 Jean-Claude Risset, Sud: Shepard-like effect using comb filters
13:00 changing the interval between components
15:00 Music example: Charles Dodge, Speech Songs
26:00 resynthesis of Dodge piece
28:00 vocal synthesis compared with phase vocoder: transposition and formants
39:00 historical example: Risset analyzing trumpet tones
43:00 evolution of sound quality in a musical note (Cake trumpet note intro)
45:00 spectrum of trumpet tone changes with dynamic ("force") 
57:00 Frequency modulation as a synthesis technique (FM synthesis)
58:00 the two frequencies (carrier and modulation frequency)
1:00:00 changing modulating oscillator amplitude changes strengths of partials
1:03:00 setting carrier and modulating frequency equal
1:04:00 spacing between partials is modulation frequency
1:04:30 making inharmonic spectra
1:07:00 using two modulating oscillators
1:11:00 applying nonlinear functions to a sinuasoid
1:12:30 why f(x) = x^2 doubles audible frequency
1:13:30 cubic and quartic polynomials
1:18:00 "between" object applies an interesting non-polynomial function


msp.ucsd.edu/media/170.21s/170-08b-may20.mp4: Meeting 16, May 20, 2021

2:00 incitement: partial tracer from Pd help browser, 4.data.structures
7:00 editing the partials in a recorded voice
11:00 historical overview of analog music synthesis
12:00 voltage controlled synthesizers
13:00 Max Mathews's MUSIC, 1957, predates voltage-controlled synths (ca. 1964)
15:00 the Telharmonium (Thadeus Cahill)
18:30 Theremin and Ondes Martinot
24:00 voltage-controlled synthesizers in more detail
29:00 music: Morton Subotnick, The Wild Bull (Buchla; San Francisco)
31:00 sequencers used in Subotnick
33:00 music: Wendy Carlos, Switched-on Bach
36:00 difference between multitracking and tape splicing approaches
40:30 how S-OB upset the boundary between "art" and "pop" music
45:00 workings of voltage-controlled synthsizers
46:30 keyboard for pitch control
47:00 "voltage control" in acoustics library objects
48:00 logarithmic control of pitch in analog vs. linear in digital
50:00 one-volt-per-octave tradition in analog synthesis
52:00 the analog gear exponentiated its frequency control inputs
53:00 module names in voltage-control synthesizers (VC oscillator, etc)
54:30 digital modules usually use linear scales such as Hz.
56:00 to combine two pitch inputs, multiply instead of adding
57:30 keyboard control and notes versus sounds
58:30 ADSR envelope (note shaper)
1:00:00 ADSR output (note: the digital one puts out curves, not line segments)
1:00:30 multiplier replaces VC amplifier
1:03:30 VC resonant filter (bandpass object)
1:05:00 sample and hold
1:06:00 keyboard control (faked from PC keyboard but should be clavier)
1:07:00 keyboard outputs: a pitch and a trigger (1 if down, 0 if up)
1:08:00 connecting keyboard trigger to ADSR
1:10:00 outptut from typical ADSR envelope in analog (made up of ramps)
1:13:00 adding another, slightly detuned VCO
1:14:30 changing Q value in resonant VCO (bandpass filter)
1:15:00 different "keyboards" Buchla vs. Moog


msp.ucsd.edu/media/170.21s/170-09a-may25.mp4: Meeting 17, May 25, 2021

0:00 incitement: making VST plug-ins in Pure Data (Camomile by Pierre Guillot)
1:30 Guillot's demo 11 on Vimeo shows how to edit Pd patches inside a DAW
3:00 the Ardour DAW (unlike Ableton - open source; intended for recording engineers)
10:30 drum pad recording as input
11:00 the Pd plug-in (Pd live patching within VST)
13:00 MIDI units to control an oscillator in Pd
13:30 managing VST parameters in Camomile
14:00 example: changing pitch and amplitude of a sinusoid using VST parameters
15:30 example: "Bedrum" plug-in
17:00 drum pad input through Bedrum patch (Irwin's test recording)
19:00 pointer to IRCAM demo showing collaboration with Irwin
20:00 the Bedrum plug-in inside Ardour (the DAW)
21:30 controlling a VST inside someone else's DAW over the network
25:30 dealing with network latency by running everything inside Irwin's machine
28:00 week's topic: sound art and sound studies
30:30 early progenitor of sound art: Luigi Russolo, the Art of Noise
31:30 introductory paragraph: (apparent) fistfights outside a theater
33:00 Russolo's instrumentrium (new mechanical noise-making musical instruments)
35:30 sounds of Russolo's instruments (Risveglio de una Citta)
38:00 non-electric (Italy in the 1920s)
41:00 later developments in novel musical instruments cosidered as sound art
43:00 Baschet brothers' instruments
44:30 Peter Ablinger and Winfried Ritsch, speaking piano
51:00 Trimpin plays Bach cello suite
53:00 Trimpin falling-water-droplet instruments
56:00 short excerpt showing a robot drumming on an old PC
58:00 sound itself as medium.  example: Janet Cardiff 40-voice motet
1:03:00 Liz Phillips, water instrument, with Joan La Barbara
1:04:00 creating standing waves activated by La Barbara's singing
1:06:00 Liz Phillips, Electric Spaghetti
1:08:00 example of visualizing sound: Chladni patterns
1:10:00 Vivian Caccuri, Ritual - candle flames set in motion by sound


msp.ucsd.edu/media/170.21s/170-09b-may27.mp4: Meeting 18, May 27, 2021

0:00 incitement: leap controller as musical instrument
9:00 crazy pool table ball trajectories
12:30 setup as duo performance with Kerry Hagan at NIME conference
13:30 strategy for controllnig instruments with hand gestures
16:30 intro to lab assignment 9: Heisenberg uncertainty principle
30:30 academic discipline: sound studies
32:00 ownership and attribution of musical sounds
24:00 Graceland album by Paul Simon with musicians from South Africa
34:30 Call Me Al, live in Hyde Park
38:30 songs were written (back in US) after recording sessions
39:30 recordings took place during Apartheid era
42:00 questions raised: appropriation or generosity?
45:00 flute solo from deep in a non-US tradition and style
46:00 Bathiki Kumolo's highly recognizable bass lick
53:00 another sound studies topic: sound recording and reproduction
54:00 Victrola phonograph
56:00 scores, "records", broadcast radio, and social media
57:00 permanance and repeatability of recordings ("records")
59:30 scores as a democratized way of sharing music
1:00:00 cost of physically distributing records and need for centralization
1:02:00 making personal copies of recorded media
1:04:00 broadcast
1:05:00 revenue model for broadcasting
1:07:00 differences between network and broadcasting
1:07:30 new revenue model for internet distribution of music
1:10:00 music itself ("content") subsidiary to network infrastructure
1:13:00 importance to artists of keeping up with changes in distribution
1:13:30 recommended books in sound studies: sound studies reader
1:14:00 Noise, Water, Meat, and Listening Through the Noise
1:15:00 architecture and its implications for music


msp.ucsd.edu/media/170.21s/170-10a-jun1.mp4: Meeting 19, June 1, 2021

0:00 incitement: making a sound installation using Raspberry Pi computer
2:00 four sound portraits, Charles Dodge tribute, built into an old telephone
3:30 getting audio into and out of a Raspberry Pi
9:00 the patch runs autonomously and sings or speaks words using a random algorithm
12:00 musical example: Paul Lansky, Idle Chatter
15:00 audio synthesis: linear predictive coding
16:00 More Lansky: Alphabet Book
18:20 techniques for machine-aided composition (compositional algorithms)
19:00 generating scores from computer algorithms
21:00 score editors versus score typesetting systems
22:00 using lilypond to set a melody from a text file with pitches
23:30 example: computer generated sequence (chromatic scale)
25:00 example: using modular arithmetic to generate circle of fifths
32:30 early compositional algorithms didn't use computer typesetting
33:00 entire computer generated musical peices versus computer as assistant
34:00 example: traveling salesman problem for chord progression (Magnus Lindberg)
37:30 feeding computer-generated pitch sequence to a modular synth
39:30 examples: modular-arithmetic clave
42:00 using the C programming language (a classic programming language)
45:00 why make efficient computer programs?
46:30 compositional algoritms in two workflows: batch or real-time
49:00 real-time text-based systems: SuperCollider, RTCmix, Chuck
50:00 modular example redone as real-time compositional algorithm in Pd
59:00 similarity to Shepard tone
1:00:30 tradeoff between text-based and graphical approaches
1:02:00 stochastic (random and pseudorandom) generators
1:04:00 re-seeding a pseudorandom number generator
1:07:00 typesetting result of a graphical compositional algorithm
1:09:30 back to Lansky.  Sampling with and without replacement
1:12:00 back to generating rhythms: changing time onset values


msp.ucsd.edu/media/170.21s/170-10b-jun3.mp4: Meeting 20, June 3, 2021

0:00 incitement: GEM, Mark Danks and Iohannes Zmoelnig, graphics for Pd
4:00 using amplitude of incoming sound to affect graphics
5:30 drawing a polygon
8:30 bird's beak is three triangles
10:00 acoustical attack detection
13:30 machine learning in compositional algorithms
14:30 data driven algorithms
16:30 perceptrons (1950s?) and back-propagation (1980s)
18:30 old-fashioned machine learning: Markov-chain Bach invention
25:00 order of a Markov chain
28:00 mixture of two Markov chains
32:00 rule-driven algorithms (as opposed to data-driven)
33:30 artificial neural networks (networks of perceptrons)
36:00 learning problem example: XOR
43:00 multiple layers in a neural network
45:30 neural networks as splines (curve or surface fitting)
48:30 example: Michael Lee and David Wessel, training instrumental timbres
54:00 limitation: trying to fit something that isn't deterministic
57:00 Judy Garland voice-controlled saxophone
1:00:00 supervised versus unsupervised machine learning
1:04:30 another example: Sam Pluta's joystick-controlled synth
1:09:00 unsupervised algotrithms: example, speech recognition
1:11:00 autoencoders
1:15:00 machine learning and time series.  Recurrent neural networks (RNNs)


-----------------------------------------------------
Music 171. Computer Music.
Winter quarter 2020.
course materials: msp.ucsd.edu/syllabi/171.20w/
recordings: msp.ucsd.edu/media/171.20w/
contents of recordings:

1a.jan7.mp4
00:00 Intro and mechanics of course
18:40 Getting Pure Data (Pd) and running it
30:36 Configuring and testing audio in and out
47:45 Editing patches in Pd
1:04:00 Organizing files so that Pd finds the out1~ helper object
1:09:30 The scope~ helper object

1b.jan9.mp4
0:00 downloading patches from browsers (patches look like text files)
5:00 Audio signals and control messages (types of connection)
6:00 Boxes: objects, messages, numbers
12:00 more on audio signals and control messages
15:41 Scope~ object to see oscillator output.  Setting frequency of oscillator
24:00 Making a complex tone with three oscillators
27:00 Tilde convention.  Multiplier (*~) as amplitude controller
35:00 Using a number box to display/change numbers
48:30 Using line~ to make continuous changes of amplitude
57:00 Messages with two numbers and "$1" convention for making variable messages
1:01:15 Messages into signal inputs and vice versa 
1:02:00 peek~ helper object to look at line~ output
1:04:00 explanation of button and toggle on scope~ interface
1:05:00 Connecting to multiple messages to set them off simultaneously
1:15:00 Osc1~ creation argument versus sending it messages
1:15:45 Osc1~ first input takes signals.  Osc1~ help window
1:19:00 Details about homework 1

2a.jan14.mp4
0:00 Wavetables (arrays of numbers)
2:35 New objects used this week
5:30 Definition of phase (with sinusoid as example)
9:30 Phasors and wavetables as components of a digital oscillator
11:30 cosine wavetable and interpolation in table lookup
22:10 Changing the range of a numerical control or signal value
24:00 Using phasor~, array, and tabread~ to make a variable-waveform oscillator
35:50 Two wavetable oscillators with slight detuning
43:50 Intro to homework 2 (array to control pitch sequence)
46:50 Web page to get patches and th flute soundfile used in homework
51:00 Creating more than one phase signal with a controllable phase difference
1:05:50 using adc~ to record sound into an array and tabread to "sample" it
1:11:20 Setting the range of lookup into an array holding a recorded sound
1:15:30 Writing and reading arrays to soundfiles using soundfiler object

2b.jan16.mp4
0:00 Interpolating and non-interpolating table lookup
5:00 Setting an array to hold a cosine cycle using "cosinesum" message to array
5:30 The send object (used to send a message to an array)
10:26 Non-interpolating lookup and how it sounds as a waveform
16:10 tabread4~ and interpolating table lookup
20:00 Cubic (4-point) interpolation explained.  N-3 available points in table
29:00 Setting range of phasor~ to (1-9) to fit into 11-point table
42:30 interpolating and non-interpolating into an audio recording
52:00 Range of phasor~ for reading into audio recording
1:02:00 De-clicking a looped read from a recording using a half-cosine envelope
1:09:30 Overlapped playback from a recording to make a more continuous sound

3a.jan21.mp4
0:00 overall topic: control and audio distinction in more detail
1:40 make a table-reading sound to control
2:10 problem to watch for: multiply defined arrays
3:40 another possible problem: signal (DSP) loops
7:00 patches on website are now zipped
8:00 introducing homework 3
11:30 control version of table read (tableread instead of tabread~)
13:40 "mod" object (modular arithmetic)  for wraparound control table lookup
14:30 comparing wraparound lookup using phasor~ (DSP correlate of above)
20:50 the "metro" obejct (metronome)
23:00 toggle and bang objects (GUI controls)
28:30 the "float" object for number storage
29:30 hot and cold inlets (left-most inlets cause actions; others modify them)
33:50 making a control-object loop (a 0-to-11 counter)
40:00 another possible problem: infinite loops of control messages
48:00 how to make presets using send and receive
57:30 semicolon-separated messages in message boxes (sending to receive objects)
1:06:30 the "mtof" object, MIDI pitches and frequencies
1:10:00 speak preview of spectrum object to see effect of frequency ratios
1:19:45 C major chord, as MIDI pitches and ratios

3b.jan23.mp4
0:00 introducing this table of contents
1:00 starting from patch from Jan 16, to work on amplitude and pitch control
5:00 showing the enveloped waveform and a simple timbre control
10:00 notes versus sounds as musical building blocks
12:00 problem with using "line" directly as an amplitude control
13:00 "delay" object to sequence line~ level changes to form notes
13:40 the bad fade-out behavior line~ gives over 10 seconds.
16:00 square root of amplitude as approximate loudness
19:00 squaring line~ output for a better amplitude ramp
22:22 introducing MIDI keyboard control
33:00 MIDI and musical intervals
39:00 why 3.86 MIDI units makes a major third (5:4 ratio)
43:00 transposing MIDI
44:00 making MIDI keyboard set off notes by triggering amplitude change
52:00 making the timbre change over the duration of a note
59:00 review of the half-cosine envelope control for the table waveform
1:04:30 printing messages that go to line~ to control it over time
1:06:00 "$1" inside message boxes to make variable messages
1:09:00 metronome-driven counter to "route" object to make rhythms
1:15:30 miscellaneous questions and answers


4a.jan28.mp4
0:00 intro to topic, voice management and polyphony using subpatches
1:00 subpatches
3:00 inlets and outlets for subpatches
5:15 abstractions versus one-off (ad hoc) subpatches
7:30 warning, don't send a 'float' to delay unless you want to set delay time
10:00 moving a sinusoid oscillator patch into another patch as an abstraction
12:40 saving an abstraction from inside the calling patch
14:00 inlets adn outlets in abstractions
18:20 commas inside message boxes send multiple messages
23:00 counter and route to distribute notes across copies of an abstraction
26:45 the pack object
28:30 the trigger object
38:00 Risset's bell instrument using an abstraction for each sinusoidal partial
41:00 throw~ and catch~ (preview)
44:00 how messages get into the partials in the bell patch
46:00 arguments to abstractions ("$" arguments in object boxes versus messages)
1:07:00 using "moses" object to split pitches among two different voice banks
1:10:00 making a sample-looping abstraction
1:13:00 building the second voice bank
1:16:00 the moses object

4b.jan30.mp4
0:00 incitement: running a patch inside an Organelle keyboard instrument
6:00 rewriting sampler abstraction to play notes using the sample's own attack
11:30 getting the size of a soundfile in samples
16:30 right inlet of phasor~ to reset the phase using a message
19:00 control computations insert themselves between audio sample computations
19:45 when you save an abstraction all the other copies might need reinitialization
22:00 looping samplers at different speeds
24:00 computing the length of a recording if it's transposed
25:00 using trigger to get a division to work when divisor changes
27:00 using float object to use a value after a delay
28:00 working piano sampler
30:00 using the two different timbres for a rhythmic effect
35:00 the "clone" object to manage copies of an abstraction
40:00 using an abstraction to reuse a patch from different calling patches
47:00 introducing homework assignment 4
50:00 "const" message to an array to clear it
55:00 sending a message to an array to set its values
56:30 about final project
1:00:30 idea: Hammond organ patch with added amplitude envelope control
1:02:00 better way to use "send"/"receive" technique for managing controls
1:05:00 "set" message to number box
1:06:00 setting send and receive names for number box
1:08:00 adding a label to the number box
1:09:00 an abstraction (setctl) for controlling a parameter and displaying it

5a.feb4.mp4
0:00 incitement: the Gem graphics environment for Pd
3:50 envelope following to use incoming sound level as a control
5:00 decibel units and attack detection (mentioned but not described fully)
6:00 drawing polygons in Gem
8:00 3-d coordinates in Gem
11:00 changing coordinates under programmatic control (by sending messages)
15:00 "rectangle" abstraction to draw a rectangle using arrays for coordinates
20:20 homework 5 intro
26:00 frequency modulation using two oscillators
31:20 spectrum of FM output (using spectrum~ helper object)
40:00 carrier frequency, modulation frequency, and index parameters
42:00 making harmonic spectra using FM
38:30 reflections and negative frequency sidebands in FM
44:30 ring modulation
50:00 ring modulating a complex spectrum
53:00 the fundamental formula for computer music (multiplying sinusoids)
58:00 using the formula to understand the sidebands from amplitude modulation
1:03 ring modulate an incoming sound from a microphone
1:09:00 nonlinear distortion (waveshaping)
1:12:00 squaring a sinusoid doubles its frequency
1:13:00 using an array as a waveshaping function
1:16 clipping (saturation) - crude example using hand-drawn wavetable
1:16:30 changing amplitude of input to waveshaping changes timbre

5b.feb6.mp4
0:00 more details on homework 4
4:00 using $ arguments inside abstraction as an index in an array
6:00 banging a float $1 object inside an abstraction
11:00 numbers to first inlet of float
13:00 commas versus semicolons in message boxes
15:00 incitement: the phase vocoder (I07.phase.vocoder.pd in the Pd examples)
24:30 intro to homework 5
28:00 messages to line~ that's already ramping interrupts ramp to start new one
33:30 symmetry in waveshaping (even and odd harmonics)
37:00 clipping using "min~" and "max~" objects
41:20 other shapes in the waveshaping array
43:00 waveshaping versus sampling as a waveform generator
46:00 adding an offset (a "bias") to teh input to waveshaping
48:00 continuously drifting bias to get a time-varying timbre
52:00 symmetry.  Change the wavetable to a pre-prepared one (cos~ object)
57:00 evenness and oddness of functions and their use in waveshaping
59:00 odd function as waveshaper
1:02:00 using cos~ waveshaper to generate even harmonics
1:03 changing bias to get sine function from cos~ object
1:07: cosine function periodicity and bias
1:09:20 digression: the "random" object used on pitch and waveshaping index

6a.feb11.mp4
0:00 incitement: sample-and-hold.  First introduce noise~ object
7:30 random versus noise~
9:00 audio decimation
14:00 one sawtooth (phasor~) sample-and-holding another
22:35 introducing "array" object (non-graphical array storage)
26:30 introducing homework 6
29:00 introducing delays using delwrite~ and delread~ objects
37:00 multiple delay taps from a single delay line
39:30 frequency response of a single delay added to a signal
44:00 phase cancellation using delays
46:00 spectrum of combined signal adn delayed copy (comb filter)
53:00 feedback with delays
1:00:00 delay loop making a tone
1:03:00 feedback gain control
1:08:00 controlling amplitude of direct signal and delays with feedback
1:10:00 changing delay times without clicks by muting output at moment of change
1:15:30 continuously changing delay time using delread4~
1:18:00 Doppler effect from changing delay time

6b.feb13.mp4
0:00 more on shifting pitch using delread4~
1:30 homework 6
4:30 making pitches using recirculating (feedback) delay as comb filter
7:00 pulsed noise for testing comb filter
8:30 pitch of output of comb filter
10:00 spectrum of feedback comb filter output
13:30 effect of integer-sample-delay restriction of delread~ (vs. delread4~)
17:00 reverberation ("reverb")
23:00 comb filter abstraction 
24:00 Pd warned before retyping into an abstraction that's been edited
29:00 oops, delread4~ doesn't take an argument to set delay time
33:00 oops, rogue out1~ object
34:00 working reverberator
37:30 pitch shifter
41:50 phasor~ driving delread4~ to make doppler shift
44:45 adc~ into rudimentary pitch shifter to show window size limitation
46:20 windowing output of delay to stop clicks in pitch shifter
48:30 cos~ object from -1/4 to 1/4 makes a good window shape
49:00 windowed pitch shifter (no overlap yet)
50:00 adding out-of-phase copy for overlapping windows
53:00 computing pitch shift from window size and phasor frequency
58:00 slope of phasor~ output
1:02:00 figuring out phasor frequency for desired transposition: f = (t - 1)/w
1:05:00 desired pitch shift in half steps (MIDI units)
1:10:00 chorus effect (mentioned, not done)
1:12:00 more about randomness (random numbers with controllable probabilities

7a.feb18.mp4
0:00 incitement 1 - beat boxer
4:00 bonk~ attack detector object
6:20 incitement 2 - visualizing data structures in Pd
12:40 homework 6 intro
14:40 followup about sampling (making notes that come out at correct pitch)
16:30 the tabplay~ object
17:00 revisiting the 4.b.sampler.pd abstraction to get correct pitch out
20:00 making new version, 7.a.sampler-voice.pd
21:00 writing sample length to sampler-nsamps that is read inside abstraction
23:00 computing the frequency a phasor~ should run at to get desired pitch
30:20 finished sampler abstraction
35:00 how filters work: comb filters and pitch revisited
40:00 the block~ object to set block size in a subpatch to make a low-pass filter
45:00 how to get unit gain at DC in a low-pass filter (moving averages)
51:03 making a high-pass filter out of the low-pass filter
54:30 the lop~ and hip~ low and high pass filters
57:00 the bp~ (briefly) and vcf~ filters
59:00 the "Q" inlet of vcf~
1:06:00 making a pulse as a test input to vcf~
1:07:00 phasor~ into pulse table to make a pulse train
1:09:00 changing pulse width changes spectrum
1:11:00 resonant filter acting on a pulse train
1:12:00 high values of Q single out harmonics; low ones give timbre changes
1:13:00 getting rid of zipper noise when sweeping resonant frequency of vcf~
1:14:00 envelope generator driving resonant frequency
1:15:00 classic note-based subtractive synthesis
1:17:30 changing pulse width of input to vcf~

7b.feb20.mp4
0:00 incitement: high-Q resonant filters for resonating ambient sounds
4:30 intro to homework 7: Devid Wessel streaming demonstration
13:00 back to subtractive synthesis
18:30 adjusting gain of vcf~ to adjust for variable Q
20:00 higher Q values make longer-lasting resonances
22:00 Q as number of wiggles in the impulse response independent of frequency
25:00 taxonomy of filters (definitions of cutoff/rolloff frequency, etc)
28:00 passband, stop band, and transition region
30:00 resonant frequency and bandwidth of a resonant filter
32:00 definition of Q
34:00 Q=17 is about a half-step wide
37:00 mass-and-spring physical analogy
40:00 how to make a resonant filter using rotations in the plane
42:00 using complex numbers to represent rotations
45:00 1, A, A^2, ... sequence is a sinusoid if A is a unit complex number
45:30 how to analyze the behavior of a recirculating delay network
48:00 impulse response of a recirculating delay network
54:00 complex sinusoid again (using Z as the ratio: ..., Z^-1, 1, Z, Z^2, ...)
56:00 analysing frequency response of recirculating delay network
1:01:00 setting up equation relating output to input
1:02:00 geometrical interpretation of frequency response
1:05:00 how resonant frequency and bandwidth depend on recirculation gain
1:08:00 "hilbert~" filter to get phase-quadrature version of a real sinusoid
1:12:00 single-sideband modulation
1:15:00 the phaser (guitar effect)

8a.feb25.mp4
0:00 about homework 7: text object will help (but isn't altogether needed)
2:00 a musical sequence as a text file
4:45 "text" object
6:00 text object subwindow is formatted like message boxes
7:00 "text" object help window.  "text define" makes and names a text object
8:00 "text define -k" to keep the text between Pd sessions
8:45 "text get"
9:30 message selectors and list messages (needs "list" if first item is symbol)
11:00 messages starting with numbers are implicitly "list"
12:00 getting single item ("atom") in the list
13:45 "text set" object to replace lines or single items in a "text"
17:00 making a simple sequencer using text object
19:00 variable time delays in sequences.  Line starts with initial delay
21:00 bad way (time value is time to next line) and good way
25:30 "text sequence" for built-in sequencer
28:00 "text sequence poodle -w 1" to make a simple sequencer
29:30 tempo control
31:30 looping sequencer using "text" object
34:00 meaning of "stop" messages to "text sequence", "delay", etc.
36:30 incremental versus accumulated time in sequencer design
39:00 polyphonic sequencing from a text file (Bach invention example)
43:00 making a synthesizer voice with controllable note lengths
46:00 "-c" flag to read text files without semicolons in them
48:00 sequencing ignoring time values
49:00 using "text sequence" to sequence the Bach example with correct times
51:00 voices and voice allocation using the "poly" object
58:00 "makenote" object to schedule note-offs after specified delays
1:02:00 new version of 8.b.voice.pd so that it takes note-on/note-off messages
1:09:00 possible variations in message style inside text sequences
1:11:30 a sequence from a real piece of music containing arbitrary messages

8b.feb27.mp4
0:00 incitement: leap controller as musical instrument controller
0:45 imaginary pool table as oscillator pair
3:30 eight parameters to control the pool-table oscillator
4:00 8 parameters derived from hand shape detected by controller
8:00 sensible controller design principle: easy on/off and loudness control
12:30 reverberation from spatial location and loudness from angle
16:00 intro to homework 8
26:00 2 miscellaneous things about filters: filtering a sinusoid does little.
28:00 low-pass filter to solve click problems
31:00 low-pass filter is smoothing in time
35:00 designing timed loops that stop at specified point
38:30 using delays to control order of execution is hard to extend or debug
39:00 loops with no internal delays using "until" object
43:00 incidentally, there's a "bang" object, does same thing as "t b"
44:00 "forloop" abstraction
48:00 you can just send a number to "until" to get that number of bangs
49:00 using loops to evaluate formulas to load into an array
49:30 example: Cauchy distribution: 1/(1+x*x)
52:00 center the curve at 51 for a 103-point table (for use with tabread4~)
56:00 Gaussian curve
58:00 "expr" object to simplify trigger-driven horrorshows to do algebra
1:02:00 the "env~" and "sigmund~" objects to estimate signal power and pitch
1:03:00 the decibel (dB)  scale (output from env~)
1:04:00 the VU meter
1:04:30 the "dbtorms" object converts decibels to linear amplitude
1:07:00 envelope-controlled oscillator
1:08:00 env~ output at 44100 Hz. appears every 512 samples (11.62 milliseconds)
1:09:00 sigmund~ pitch tracker (and more)
1:12:00 voice-controlled oscillator (pitch and amplitude)
1:13:00 quantizing (rounding pitch off to nearest integer)
1:15:00 starting on presets (wil finish in 9a video)
1:17:30 "presetter" abstraction will be setctl plus preset saver
1:19:00 printing out the name of the preset variable using "symbol" object

9a.mar3.mp4
0:30 finishing the presetter abstraction
2:00 pack and unpack with symbols
4:30 message selectors (automatic only if list starts with a number)
5:00 "bang", "float", "symbol" and "list" selectors
7:30 printing parameter messages returned from presetter
8:00 collecting them in a "text" object ("text set" to append lines)
11:00 clearing the text object out before parameters are put in using trigger
16:00 incitement: waveshaping (distorting) combined sounds of different periods
19:00 clip~ as waveshaper
20:00 designing smooth waveshaping functions
22:00 y = x - x^3/3 cubic curve as waveshaper
27:00 cubic waveshaper without clipping
29:30 clipping the input to the cubic
32:00 sum of 440- and 660-hz sinusoids into waveshaper gives common subharmonic
36:30 oops, we were listening to the wrong signal before - fixed now
40:30 another incitement: deliberate unstable feedback through a delay line
42:30 test signal into a pitch shifter
44:30 soft-clipping the output of the pitch shifter and feeding it back
49:00 feedback dies if it's pitch shifted out of range
50:00 restricting the frequency range of feedback with hip~ and lop~ filters
54:00 changing the pitch-shift delay time and frequency range
1:00:45 incitement: measuring tempo using "bonk~" and "timer" objects
1:05:00 trigger to control two "bang" messages to "timer"
1:06:00 using that to drive a "metro" object
1:07:00 controling a test beep from the result
1:08:00 triggering sample recording from bonk~ and playback from metro objects
1:12:00 foldover (intro for weeks' topic, band-limited waveforms)

9b.mar5.mp4
0:00 incitement: partial tracer (Pd audio example)
11:00 sinebag example - real-time additive analysis/resynthesis using sigmund~
14:00 back to foldover issue.  Pd audio example demonstrating foldover
18:00 making a band-limited sawtooth wave
23:00 approximating a sawtooth summing sinusoids
28:00 square wave as sum of sinusoids
31:30 using the square wave as transition for a step in an audio signal
34:00 oscillator repeating the step
35:00 changing the duration of the ripple portion of the step
37:00 subtracting the original phasor~ to get rid of the discontinuity
39:00 changing the band limit on a sawtooth wave
40:00 comparing the three sawtooth generators (phasor~, fixed partial sum, and new version)
42:00 spectrum of band-limited sawtooth
45:30 making a rectangle wave out of two sawtooth waves
48:00 why subtracting two sawtooth waves gives a rectangle wave
50:00 spectrum of rectangle wave
54:00 triangle waves out of phasor~
57:00 spectrum of symmetric triangle wave
1:00:30 non-symmetric triangle
1:04:00 triangles as difference between two parabolic waves
1:06:30 spectrum of parabolic wave
1:08:00 parabolic waves subtracted
1:08:30 trapezoidal wave: sum of parabolic waves with gains that sum to zero

10a.mar10.mp4
0:00 prep for final - making screen movies (I use ffmpeg; OBS easier to use)
5:30 how to turn in final projects
8:20 incitement: video feedback using Gem
10:30 making ribbons using hundreds of triangles
11:00 looking up colors and coordinates from tables (arrays)
11:30 making the snake slither by sliding forward in the arrays
13:30 making video feedback in Gem: pix_snap2tex object
15:30 reducing size and orientation of fed-back copy
17:30 adding a second feedback path
19:20 topic: frequency domain processing using example from Pd doc
20:10 The J series of Pd audio examples are on previous class's topic
21:00 The I series, recalling the phase vocoder incitement from earlier
21:30 Motivation: Problems with time streching via granular sampling
23:00 Problem of overlapping audio recordings without phase cancellation
24:30 idea of Fourier analysis/synthesis.  Path I01.Fourier.analysis.pd
28:30 Fourier analysis period N controlled by block size in Pd
30:30 typical block sizes are 512-4096 but first example uses 64 samples
34:00 phase of test oscillator is reset each window
36:00 different initial phases to input of Forier analysis
36:30 explanation of fft~ output as real and imaginary parts of analysis
37:10 64th point of output is same as zeroth point
38:00 complex exponentials as true basis functions (instead of sine/cosine)
40:20 cos(phase) + i sin(phase).  Phase is 2 pi n / N for fundamental
41:00 negative frequencies. Real-valued cosine is sum of two complex exponentials
43:00 complex sinusoids are eigenvectors of time
45:00 complex multiplication used to knock frequemcy down to zero for analysis
47:00 how initial phase of real sinusoid changes fft~ output
49:00 real and imaginary part of tenth fft~ analysis graphed in complex plane
51:00 analyzing sinusoid with frequency 10.5 (between harmonics)
54:30 partials drop off as 1/k from peak
56:00 using a raised cosine window as (Hann) windowing function
57:00 using windowing function helps resolve sinusoidal components
58:00 example I02: applying Hann window function
59:00 the tabreceive~ object.  Setting window size to 512 via block~ object
1:00:00 the rfft~ object and taking magnitude
1:01:00 changing window function.  Heisenberg uncertainty principle.
1:05:00 filtering using Fourier analysis/resynthesis
1:07:00 overlapping windows 
1:09:00 applying windowing function both before and after analysis/resynthesis
1:11:00 analysis/resynthesis patch with blocking, windowing, and normalization
1:14:00 fft vocoder, patch I06.timbre.stamp.pd
1:17:00 changing pitch of voice
1:18:00 how the patch works

10b.mar12.mp4
0:00 incitement: drum machine using PAF (fancy waveshaping) synthesis
3:00 audio production techniques
4:00 dynamics processing: companding compression and expansion) and limiting
15:00 filtering, DJ-style, using high-order (Butterworth) filters
16:30 returning to Fourier analysis/resynthesis
18:00 analysis/resynthesis considered as filterbank
20:00 arranging phase coherency in each channel in resynthesis
25:00 quick tour of I07.phase.vocoder.pd patch
26:30 topic: panning and cross-fading
29:30 what gain should we use half-way across a pan or cross-fade? 1/2 or 1/sqrt(2)?
32:40 How pan should depend on left/right control (not linearly)
36:00 quadraphonic speaker setups for 1950s - 1970s style spatialization
42:00 readsf~ and writesf~ to stream from and to soundfile in real time
48:00 sending messages over networks
50:30 messages between Pd and other programs: pdsend and pdreceive
57:45 conditionals in Pd: route, moses, spigot
1:00:00 reminder, execution order
1:01:00 control messages from signals: env~, snapshot~, threshold~
1:05:00 state machine to detect audio onsets


-----------------------------------------------------
Music 270c. Compositional algorithms.
Winter quarter 2021.
course materials: msp.ucsd.edu/syllabi/270c.21w/
recordings: msp.ucsd.edu/media/270c.21w/
contents of recordings:

msp.ucsd.edu/media/270c.21w/270c-01a-jan5.mp4: Meeting 1, Jan. 5, 2021

0:00 overview of syllabus
6:40 assignment 1
10:15 solution 1: using pure data
13:00 pure data utility patch for playing and seeing musical notes: 0.common.pd
14:00 Pure Data patch library
16:15 Pd "mod" object (integer modulus)
21:50 circle of fifths as 7n mod 12
24:00 solution 2: using julia and csound
27:00 printing csound "note cards" in julia
34:30 julia solution requires a rewrite of the math to vectorize it
37:00 relationship between pitch (in MIDI units) and frequency
38:00 major and minor thirds in common and graphical notation
40:00 pitch continuum ("floating point MIDI units")
41:00 just and tempered triads as pitch
43:30 conversion from pitch to and from frequency (mtof and ftom objects)
51:30 overtone series in musical notation
54:00 frequency shifting (as pitches and freqencies no notes)
57:00 odd harmonics of A55
1:01:30 inharmonic frequence shifts
1:03:30 intervals between pairs of notes with fixed frequency spacing
1:07:00 pitch related to log frequency (midi = 69 + 12*log_2(frequency/440)
1:09:00 negative frequencies.  Frequency shift to generate microtonal pitch sets
1:10 callout to Stock, Mantra and Chowning, Stria as examples
1:12:30 quantization.  Rounding to nearest integer in Pd
1:20:00 next time: non-integer-spaced pitches rounded to integers


msp.ucsd.edu/media/270c.21w/270c-01b-jan8.mp4: Meeting 2, Jan. 8, 2021

0:00 Continuing with pitch manipulation: Western scale as int(12n/7)
2:00 the "for" object in the 270c library, counting to 7
11:00 adding 1 to change western-scape mode
22:00 non-octave scales, genral maximally uniform spacing
23:30 specifying time onsets
25:00 maximally-uniform time onsets
26:00 quintuplets, 5 against 4 against 3
27:30 quantized rhythm
30:00 western scale as rhythm
33:00 repeating the quantization process
38:00 western major scale as 3 triads end to end, also maximally uniform
41:00 building major scale as f-a-c-e-g-b-d
43:00 Balzano picture of 7-tone scale in 12-tone octave
43:45 more operations on pitches. Quantizing to just scale
50:00 pitch sequence to try
51:00 text object for searching through list of pitches
59:00 general tool for quantizing to a set of pitches
1:01:00 interpolating between pitch sets
1:03:00 preliminary: stretching and squeezing interval sets
1:04:00 inversion
1:05:00 sliding through different stretch factors (using trigger object)
1:06:30 back to interpolation
1:08:00 the map object in the 270c library
1:09:30 expression for weighted mean (aka linear interpolation)
1:10:00 oops, need a trigger object
1:15:00 oops, rounding negative numbers works differently
1:16:30 working interpolation
1:17:30 extrapolations

msp.ucsd.edu/media/270c.21w/270c-02a-jan12.mp4: Meeting 3, Jan. 12, 2021

0:00 generalities about formalized composition
3:00 Xenakis sets a piano on fire
5:30 folding via modulus and absolute value
7:00 revisit assignment 1 using folding
11:00 interval between 2 pitch classes 
17:00 mod 12, minus 6, absolute value, plus 6
19:30 interval vectors
20:00 serialize (etc) objects in library
22:30 interval vector of (0,3,6,9) chord
24:00 generatING all pairs from (0,1,2,3)
30:00 map the pairs through the chord to get all intervals
33:00 music theory: structure of complete tetrachords
38:00 how to complete an interval into a complete tetrachord
39:00 example - chains of tetrachords sharing bichords
43:30 algorithmic search for chords containing subsets
44:00 nested for loops in Pd
48:00 counting with the second digit as least significant digit
50:00 choosing each possible pair of digits once
54:00 finding all the intervals in a tetrachord using "for" loops
56:00 serialize and parallelize library objects
59:00 tohisto and fromhisto library objects
1:05:00 finding library objects and documentation in 0.common.pd
1:07:30 prepare for how to find all chords containing a set: set intersectios
1:11:30 apply library object
1:13:00 iterate through transpositions of a chord
1:17:00 intersection of a fixed set with transpositions of a second one
1:20 finding the unique E/F# bichord in a transposed (0 5 6 8) tetrachord

msp.ucsd.edu/media/270c.21w/270c-02b-jan15.mp4: Meeting 4, Jan. 15, 2021

0:00 interpolation between sequences
0:30 text object to hold a sequence
4:00 reading sequence using "text get"
12:30 incorrect sequencer
15:30 playback tempo control
17:00 text editing wizardry failure
20:00 corrected sequencer
21:50 same but with expr so that tempo control isn't upside-down
22:30 get notes out of both sequences in parallel
23:45 Parker rhythm applied to Bach pitches
24:30 pitch squeezing/stretching using expr
28:00 inverting about half-integer pitches
30:00 pitch interpolation between sequences
33:30 pitch interpolation plus stretch/squeeze
35:45 interpolating the rhythms (inter-onset intervals and durations)
45:00 application: P/H Concrete, Xenakis
46:00 straight-line paths through a hyperbolic paraboloid
49:00 pairs of P/Hs sharing edges
52:00 same structure as time/frequency pairs
57:00 rhythmic interpolation in Jupiter by Manoury

msp.ucsd.edu/media/270c.21w/270c-03a-jan19.mp4: Meeting 5, Jan. 19, 2021

0:00 assignment 2: 12-bar blues noodler
3:00 probability distributions
4:30 the "random" object in Pd
5:00 pseudorandom number generators and seeds
7:00 Restarting Pd resets random seed.  (Deterministic stochastic process)
10:30 a simple pseudorandom number generator
14:00 histograms to measure probability distribution
21:00 checking flatness of probability distribution coming from "random" object
25:00 histogram of digraphs
30:00 small number of trials gives crazy histograms
32:00 thought experiment: write a piece of music in which everything is random
37:00 Xenakis's idea of stochastic music
40:30 what the seed does internally
42:00 generating a random number with a given probability distribution
54:00 verifying the probability distribution with a histogram
1:00:00 interpolating bwtween two distributions
1:01:00 quantiles as generalization of random choice with given distribution
1:05:00 interpolation version 1: cross-fading between distributions
1:11:00 interpolating horizontally between two distributions

msp.ucsd.edu/media/270c.21w/270c-04a-jan26.mp4: Meeting 6, Jan. 26, 2021

0:00 more on randomness: sampling without replacement
11:00 arpeggiation without repetition (Lansky technique)
17:00 random permutations
20:00 example: voicing chords as permutation
36:00 random processes that depend on previous outcomes
39:00 stationary processes
42:00 Markov chains versus memoryless processes
47:00 memory of future outcomes?
52:00 Markov condition: dependence on past is limited to immediate past
54:00 random walks (example of a Markov chain)
1:06:00 making a Markov chain in Pd
1:07:00 higher-order Markov chains
1:08:00 limitations of Markov modeling of musical sequences

msp.ucsd.edu/media/270c.21w/270c-04b-jan29.mp4: Meeting 7, Jan. 29, 2021

0:00 training a Markov chain from a corpus
4:00 sparseness of Markov chain matrices trained from a corpus
5:30 continuous versus discrete state spaces
8:30 alternative to building a Markov transition matrix: indexing the corpus
11:00 example: smerdyakov object, used by Manoury
16:00 encoding extra information when states are labeled by pitch
20:00 polyphonic music through smerdyakov

msp.ucsd.edu/media/270c.21w/270c-05a-feb2.mp4: Meeting 8, Feb. 2, 2021

0:00 more on random walks: boundarys at end of finite-length walks
5:00 adding stationary transitions to a random walk
6:00 boundaries on states or between states
7:30 implementation in Pd
13:00 histogram showing steady state probabilities of random walk
16:00 adding gravity (asymmetric step probabilities)
17:00 adding stationary transitions in Pd
20:00 imitating continuous time (discrete state) Markov process
21:00 diffusion rate of the process
26:00 tour of smerdyakov c code
30:00 mixtures of two Markov chains
32:00 oops, 4th order, no transitions between Donna Lee and Bach invention
33:00 weighting the mixture to crossfade between the two
37:00 don't play transitions that have long delays
39:00 back to c implementation
42:00 methods of smerdyakov object
43:30 "tick" method
44:30 how variable-order Markov chain is implemented
47:00 adding up weights of all transitions (used later to normalize)
48:00 uniformizing rhythm
54:00 writing out a random walk explicitly in smerdyakov
57:00 chord slop (playing back chords that are recorded from a keyboard)
59:00 text-based Markov chain as demonstration of adaptively variable order
1:06:00 varying the order as function of desired number of continuations
1:10:00 entropy
1:12:00 incitement: when is one ramp greater than one at a different increment?

msp.ucsd.edu/media/270c.21w/270c-05b-feb5.mp4: Meeting 9, Feb. 5, 2021

0:00 deterministic processes with desired statistics (z12 process)
12:00 keeping track of each state's excess
16;00 dueling metronome (equivalent process)
19:00 outputs a familiar sequence (01001010...)
20:00 l system that outputs the same thing (actually its complement 10110101...)
21:30 incitement: convolving 1000 and 1618 msec metronome outputs
27:00 same thing as sweeping through an orchard (lattice points in first quadrant)

msp.ucsd.edu/media/270c.21w/270c-06a-feb9.mp4: Meeting 10, Feb. 9, 2021

0:00 supervised and unsupervised machine learning
3:00 artificial neural networks (ANNs)
6:00 perceptron function: sigmoid function of a linear combination of inputs
9:30 behavior for large inputs
10:00 xor function as thing a one-deep perceptron network can't "learn"
12:00 hidden layers of nurons / perceptrons
14;45 training the ANN as fitting a spline to input data
16:30 ANN package for pure data
18:00 training ann_mlp object in Pd
20:00 training set (file train-xor.txt)
23:00 running the trained ANN 
25:00 the trained neural net as a text file
34:00 (tried and failed to use julia to plot ANN function)
37:00 run-network-to-file patch runs an ANN so we can graph the behavior
44:00 (give up trying to graph and go on): how ANNs fail
45:45 Lee and Wessel paper, ANN to train additive synth from sound recording
52:00 subpatch playback.pd works on a single partial
55:00 training set from saxophone solo recording
57:00 is it better to train one ANN for all partials, or train separate ones?
1:00:00 testing the ANN we trained on teh saxophone
1:08:00 listening to output of ANN imitating the sax (using same input data)
1:10;00 why this is an ill posed problem for an ANN
1:14:00 2D should be too low dimensionality for this to work

msp.ucsd.edu/media/270c.21w/270c-06b-feb12.mp4: Meeting 11, Feb. 12, 2021

0:00 sonic challenge 3: choice 1: make an algorithm to attack sax solo problem
1:00 choice 2 (discrete-event problem): make a melody given a melodic contour
4:30 theoretical discussion of back-propagation training for ANNs
6:30 training as a minimization problem
9:00 total error function to minimize
10:00 optimization assumes we have a trusted distance function on the output
14:00 training in practice using ann_mlp object in Pd to learn a parabola
15:30 graphing output of ANN
18:30 graphng XOR ANN result
20:00 training for XOR again, failure case
23:00 badly trained ANN wasted 3 hidden neurons solving the same part of problem
31:00 behavior outside of training set: sigmoids better than polyynomials
37:30 ANN trained on parabola levels off outside training set
38:00 training failure example on parabola
40:45 another failure mode
43:00 surprise: adding hidden neurons doesn't make training much slower
48:00 2-dimensional paraboloid, 10 hidden neurons, graphing outside training set
50:00 successfully trained paraboloid, behavior outside training set
56:00 heatmap view
57:30 existence of python machine learning packages (not available in julia)
58:30 how to prepare inputs to ML algorithm from audio
1:00:00 is "deep learning" musically useful, either now or potentially?

msp.ucsd.edu/media/270c.21w/270c-07a-feb16.mp4: Meeting 12, Feb. 16, 2021

0:00 Sam Pluta guest talk: training neural nets to generate synthesis parameters
1:00 training on tiny training sets
2:00 inspired by Rebecca Fiebrink's Wekinator
2:30 instead of using complex inputs, try simple inputs and complex outputs
3:30 training ANNs when there's no right answer
4:00 Pluta's instrument
5:00 two x/y pads as input
6:30 two surfaces, each with 4 synth algorithms, 8 ANN mappings
10:30 training the ANN with 6 points
12:00 testing the network on one of the 6 training points
13:00 another synth, FM7 (DX7 imitation in supercollider)
14:30 Following Frederick Olafsen's example of how to use FM7
16:30 code for parametrizing FM7
17:00 controlling range of parameters
19:00 analog systems are more forgiving of unstable networks than digital ones
20:00 in FM example, controlling oscillators that then control synth parameters
20:30 how to train the network
21:00 example of a fragile network (small usable parameter range)
22:00 adding a point and retraining
25:00 classifiers don't make much sense in this environment
25:30 very narrow space that works, most parameter combinations will be lame
28:00 4-dimensional space is already impossible to visualize
28:30 impossible for a human to design the mapping the ANNs make automatically
30:00 the ANN can exactly hit the training points; fun thing is interpolations
31:00 using FLUCOMA  MOP regressor from Huddersfield
31:50 training is better in Keras but it's much slower to train or load
34:00 we're in a moment where new ideas are opening up
35:00 in this approach failures are interesting
37:00 FLUCOMA is out already
38:00 not an example of overfitting
40:00 about overfitting.  Controlling by minimizing total variation
46:00 good and bad range of output of sigmoid
48:00 best if lots of useful outputs in middle of range of sigmoid
49:30 Pd ann_mlp (based on FANN package) defaults to (-1, 1) interval
50:00 implication for training partial amplitudes
53:00 how this works out for Lee/Wessel technique
59:00 if natural output range is (-1,1) quiet partials work badly
1:03:00 normalizing before presenting trainign set to ANN
1:06:00 badly normalized ANNs on Weeks solo sounds like Hammond organ
1:12:00 theoretically Lee/Wessel ANN should never make foldover
1:13:00 Q&A and discussion
1:19:00 quick review of how to operate Lee/Wessel patch

msp.ucsd.edu/media/270c.21w/270c-07b-feb19.mp4: Meeting 13, Feb. 19, 2021
0:00 sound challenge 3 again: making melodies with desired contours
1:00 rule-based composition as optimization: melodizer.pd
4:00 digression: debugged version of sax Lee/Wessel technique
7:00 vocal recording through saxophone transform
10:00 graph of first partial amplitude as function of pitch and loudness
12:30 comparing ANN output with measured amplitudes
13:30 graph of measured tenth partial amplitudes
15:00 compared with ANN output
16:30 unsupervsed learning algorithms (no coordinate distinguished as "output")
18:30 self-organized maps (SOMs)
20:30 example: Henrik Frisk SOM analysis in free improvisation
21:30 "learning" a circle
24:00 standard SOM algorithm described
33:30 convergence of SOM algorithm
36:00 how learning parameters affects convergence
41:00 letter "phi" shape (harder problem)
44:00 topological condition for SOMs to work
45:30 using SOM output
47:30 necessity of trusting the distance measure (metric)
49:00 possible metric for spectra
53:00 code for SOM object in Pd
56:00 k-means clustering algorithm
1:07:30 example of a bad outcome 
1:08:00 outcome depends on (random) initial state
1:12:00 bad outcomes are local minima that aren't global


msp.ucsd.edu/media/270c.21w/270c-08a-feb23.mp4: Meeting 14, Feb. 23, 2021

0:00 will cover paper by Arjovsky et al., Wasserstein Gan in week 10
3:45 today's subject is optimization algorithms applied to musical composition
4:30 amusement: applying Lee/Wessel to Judy Garland vocals
8:30 separate networks work better than one combined one
10:30 Garland analysis file, 10.66 msec per slice
11:15 getting signal power (add partial power to partially reject orchestra)
13:30 melodizer: optimizing melody generator
14:00 how to import C code into Pd using "cfc" object (simpler than externs)
16:00 cfc example: greatest common divisor (gcd) using Euclid's algorithm
18:00 melodizer setup: 29-note melody with quarter-note rhythm
18:30 First term in melody's "value" or "score" evaluation: transitions
22:30 overall score is weighted sum of scores by the different criteria
23:00 criteria are: transitions, repetitions, contour, harmony, and similarity
23:30 combinatorial optimization algorithm: change 1, 2, or 3 pitches per step
25:30 even if a change doesn't improve score, make it if score doesn't go down
27:00 melody optimized according to transitions criterion alone
29:30 repetition criterion: fragments reappear.  (wrong, corrected 45:00)
30:30 Markov analysis or other local methods won't capture this
32:00 first argument to cfc message selects 1 of 5 functions
33:00 tour of the cfc source code
42:45 "getscore" c function in cfc to make weighted sum of component scores
43:45 "perturb" function alters melody
44:30 functions for the individual criteria
45:00 repetitions criterion punishes us for over-repeating pairs of notes
45:30 contour criterion
46:45 tonality criterion
50:00 similarity criterion - repeated fragments
52:00 optimizing by contour alone
54:00 (fixing a mistake in contour code and trying again)
56:30 adding transitions criterion to contour
58:00 all criteria combined
58:30 optimization strategy: relaxation (temporarily drop one requirement)
1:03:00 finished product (sonic challenge 3, melodic version)
1:09:00 Stockhausen BBC lecture: "if this were an octave the piece would be over"
1:11:00 another possiblitiy: duelling Markov chains (Baysian networks, sort of)

msp.ucsd.edu/media/270c.21w/270c-08b-feb26.mp4: Meeting 15, Feb. 26, 2021

0:00 Approach to sonic challenge 3 using conditioned Markov chans
1:00 input to smerdyakov.  Example: top line of Bach invention
2:00 changing order of Markov chain
3:30 imposing a priori probabilities
4:30 probabilitiy space representation of conditioning a Markov chain
7:00 conditioning is just hacking a smaller piece out of the sample space
11:30 Bayes' rule (flipping conditional probability backward)
16:00 specifying an a priori probability distribution
17:00 model: melodic contour is considered as blurred output of the Markov chain
19:00 Bach with random accidentals
20:00 force that (stupid) output to be a melodic contour we specify
31:00 result: just multiply transition probabilities by contour weighting function
34:00 normalization is confusing but we can just assume it's constant
36:00 message to smerdykov to impose an apriori probability distribution
38:00 testing smerdyakov with imposed contour
40:00 possible to give an impossible condition (0/0 probability)
41:00 no sanity check, but give everything an a priori prob of at least 10^-9
43:00 setting order to 0
44:30 implementation in smerdyakov source code
48:00 example of a Markov chain to generate syncopated rhythm
56:00 limitation: the three processes are ignoring each other
57:00 because of syncopation the chains don't advance at same times
58:00 for the chains to condition on each other we might need to add states
1:00:00 conditioning Markov chains on each others' (not yet computed) outputs
1:01:00 I think this is a Bayesian belief network (that's a buzz-phrase)
1:02:00 Bayes' law allows us to rearrange the arrows of causality
1:05:00 if not in real time you can re-imagine this as an optimization problem

msp.ucsd.edu/media/270c.21w/270c-09a-mar2.mp4: Meeting 16, March 2, 2021

0:00 typical rhythmic constraint: avoiding collisions in time
2:00 example from Charlotte Truchet paper: rhythmic canons
4:00 random walk, smaller time steps limit to Brownian motion
5:30 parameter of Brownian motion: variance per unit time
6:00 doubling speed multiplies standard deviation by sqrt(2)
8:00 fractal property: re-scaling speed gives another one renormalized
9:30 another fractal: snowflake curve
10:00 length is infinite
13:00 this mirros composers' idea of having small scale reflect large scale
15:00 Stochhausen's idea of playing Beethoven's 9th symphony at 440 Hz
16:00 back to L system to make self-similar sequence of 0s and 1s
17:00 there's also a top-down way of seeing it that goes from large to small
21:00 fractally partitioning a line segment.
22:00 This shows up in Poeme ELectornique,
25:00 Cyril Henry's (I think) L-system examples in Gem
33:00 chaotic flows on 4-sphere

msp.ucsd.edu/media/270c.21w/270c-09b-mar5.mp4: Meeting 17, March 5, 2021

(spliced; part of original is missing)
0:00 relationship between stochastic processes and chaos
0:30 iterating the function 4ax(1-x)
4:00 a=1: Values near zero repelled exponentially
5:00 smaller values of a. a>1/4: zero (unstable), and another fixed point
8:00 when slope of crossover point less than -1, other fixed point is unstable
10:00 period doubling
11:00 understanding the function as taffy-pulling
12:00 a=1: Cauchy polynomial - doubling the angle, so equivalent to 2x mod 1
14:00 start with a random real number, and output binary digits as a sequence
15:00 so iteration (if a=1) is equivalent to a coin-tossing process
18:00 continuous-time chaotic process
19:00 flows in in one or two dimensions
20:00 one-dimensional flows have simple behaviors
23:00 oscillators as two-dimensional flows
26:00 possible behaviors in neighborhood of fixed point in two dimensions
30:00 three dimensions or up: possibility of chaotic behavior
33:00 example: Lorenz attractor
36:00 output as audio signal
38:00 if you can't observe it exactly, you can base a random process on it
39:00 you could measure the entropy of the random process
39:30 how to make fractal curves out of a dynamical system
40:00 snowflake curve interior as eventual fate of an iterated function

msp.ucsd.edu/media/270c.21w/270c-10a-mar9.mp4: Meeting 18, March 9, 2021

0:00 example of a (non-chaotic) flow: differences between three phasors in pairs
2:00 which of two phasors is "ahead" of the other measured against a third one
4:00 same thing as a control operation (incitement from end of Feb. 2 class)
5:00 the 0100101... L-series if one is golden section faster than the other
7:30 why did subtracting 2 phasors into a comparator give 3 pitches?
24:00 state space of two oscillators can be drawn as a rectangle
27:00 flows in a two-dimensional phase space that is topologically a torus
33:00 coupled oscillator pairs
34:00 can require that all arrows in the flow point northeast to avoid stable points
35:00 flow described as a differential equation determined by a vector field
37:00 first-order equation implies no inertia
39:00 hard-syncing one oscllator to another
42:00 hard sync as a flow
44:00 soft sync
48:00 making a sawtooth oscillator from scratch in Pd (block 1 subpatch)
53:00 soft-syncing two of them using a comparator
58:00 other type of coupled oscillator network: non-physical pool table
1:00:00 simplified phase space is 4x the top of the table
1:01:00 more correctly, the phase space is four-dimensional
1:02:00 very incorrect pool table: triangle
1:03:00 assuming fixed speed can reduce phase space to 3 dimensions
1:04:30 resulting behavior
1:05:00 very strange behavior if triangle has one very small angle

msp.ucsd.edu/media/270c.21w/270c-10b-mar12.mp4: Meeting 19, March 12, 2021
0:00 topic not covered yet: coupling Markov chains to each other
3:30 another uncovered topic: flows on 3-sphere (unit quaternions)
6:30 other topics: Gendy (Xenakis); signal-rate staircase; 3F
9:00 Gendy algorithm: waveforms made of end-to-end connected line segments
13:00 Pd patch to imitate Gendy
14:00 random-walk abstraction.  Random walks with constraints and bias
21:00 S709 by Xenakis - you can check on waveforms in the original recording
23:00 Audio rate staircase: phasors and sample/hold
31:00 bell designer (3F technique)
35:00 2- or 3- dimensional lattice of frequencies
36:30 ways of choosing finite subsets of lattice
38:00 the zero-frequency line (or in general, hyperplane)
40:00 frequently occurring frequency pairs separated by f, g, or h
41:00 can be realized as oscillators or resonant filters for example
43:00 seeding the pseudo-random sieve

msp.ucsd.edu/media/270c.21w/270c-11a-mar16.mp4: Meeting 20, March 16, 2021

0:00 Arjovsky, Chintala, and Botton, Wasserstein GAN paper
2:00 Think of unsupervised learning as learning a probability distribution
3:00 ground truth is a probability distribution 
4:00 we wish to map a simpler distribution to the "real" one
4:30 insight 2: the Kullback-Leibler divergence isn't an ideal measure
5:30 definition of KL divergence
7;00 think of divergence from f to g  as "how well you can encode f using g"
8:00 why it's always nonnegative
9:30 problem with KL as a thing to minimize in a training procedure
12:30 maximizing likelihood of observations given the model
13:00 but what if likelihood is zero? adding noise to make log likelihood finite
15:30 asymptotically, measured KL divergence becomes log liklihood plus constant
16:30 added noise typically 10% of amplitude of signal
19:00 earth-mover distance.  We saw a form of this earlier (quantile cross-fade)
21:30 probably related to the optimal transport problem
22:30 hard to estimate earth-mover distance in higher number of dimensions
23:30 even NP complete problems are often tractable in practice
24:30 a minimax principle relating earth-mover distance to an integral
24:30 (Kantorovich-Rubinstein duality)
30:00 Can treat this as a supervised learning problem
31:30 maybe this doesn't measure earth-mover distance but it seems to work anyhow
34:00 general setup for auto-encoders
37:00 applying earth-mover's distance to train it
38:00 training algorithm as a nested loop (inside loop learns earth-mover distance)
40:00 this is "algorithm 1" in paper
41:00 one trick: don't have to update earth-mover function to estimate slopes
43:00 typical application: sampling from the "learned" distribution
45:30 how it's possible to affect specific choise of sample
48:00 comparison: self-organized map solves a different minimization problem
49:00 idea: make a joystick that travels through a set of desired outcomes
50:30 this might be what is called normalizing flows
51:30 IRCAM paper: train a VAE to learn how to parametrize a synthesizer
53:00 big problem area: dealing with time behavior as a learnable feature
-----------------------------------------------------
Music 271b. Survey of Electronic Music Techniques II.
Spring quarter 2021.
course materials: msp.ucsd.edu/syllabi/271b.21s/
recordings: msp.ucsd.edu/media/271b.21s/
contents of recordings:

msp.ucsd.edu/media/271b.21s/271b-1a-mar31.mp4: Meeting 1a, March 31, 2021

0:00 the electronic music repertory
1:00 part 1: early computer music
5:00 part 2: audio recordings as medium
7:00 the tape music era (1943-2000 perhaps)
8:30 part 3: algorithmic music
16:30 Outliers: Wendy Carlos and Maryanne Amacher
19:30 organization of the course (send me e-mail about what interests you)
23:00 Patches to use: the Pure Data Repertory Project: 
25:00 PDRP documentation
28:30 Example: running patch for John Chowning, Stria
31:30 about the Stria patch (not a realization, just a study patch)
32:00 FM (frequency modulation) synthesis
33:00 sinusoids and FM sound at A440
34:30 FM ratio (modulation frequency to carrier frequency)
37:00 FM spectrum
37:40 main out and mute controls.  Gain units for main out
39:30 spectrum of very slowly changing frequency tone
40:00 index controls range of frequency deviation.  Phase modulation
40:30 if using phase modulation, index adds bandwidth
42:30 frequency resolution of ears and of FFT-measured spectra
43:30 bandwidth (spread) is product of index and modulation frequency
44:30 faster modulation gives distinct spectral peaks
47:00 modulating frequency controls separation of peaks but not amplitude
48:00 index changes the amplitudes but not the frequencies
49:30 when index is 0.38 center frequency drops out.  (it bobs in and out)
51:30 using simple fractions as FM ratios (example, 1/4: periodic tone)
54:00 golden ratio as FM ratio (non-periodic)
56:00 spacing between peaks (reflection of negative frequencies)
56:00 Helholz theory of consonance and dissonance for periodic tones
57:30 musical fifth (7 half tones)
59:00 beating between common partials that almost coincide
1:02:00 making the interval sour
1:03:30 filtering to hear the individual harmonics
1:04:00 caution abut gain while filtering sounds
1:05:00 perfect fifth (7.02 hald steps)
1:06:00 tritone (dissonant interval: pairs of partials within a critical band)
1:08:30 tempered major triad
1:10:00 near-perfect triad (-3.16 and -7.02 half-steps)

msp.ucsd.edu/media/271b.21s/271b-1b-mar31.mp4: Meeting 1b, March 31, 2021

0:00 Stria by John Chowning
1:00 first tone in stria, imitated using the patch
2;00 beginning of real piece
6:30 time proportions in the piece
9:00 the pre-stored sequence in the patch
10:00 more faithful reconstruction from Huddersfield
12:00 theory: using the golden ratio as an FM ratio
14:00 geometric construction of golden ratio
17:00 r-1, 1, r, and r+1 in geometric progression (same interval 3 times)
22:00 Fibonacci sequence and golden ratio related
24:30 spectrum of FM with golden ratio of modulation to carrier frequency
25:00 stacking three golden FM spectra
26:00 musical scales with different octave ratios and octave divisions
28:00 spectrum of four tones separated by golden ratio
29:00 three golden ratios sound almost like 2 octaves
32:00 same interval is less sour if we use golden-ratio spectra
34:00 spectrum of four tones separated by golden ratio sounds consonant
37:00 1/3 of an octave (if an octave is golden ratio)
40:00 major tetrad in 12-tone scale, sounds sour if using golden ratio spectra
41:30 (0,3,6,9) tetrad plays role of major tetrad in Stria
42:30 dividing 2:1 octave in other numbers of steps, playing (0,4,7,12) chord
43:00 quarter tones (2:1 divided in 24, or tritone divided by 12)
48:30 how many steps to divide "octave" into
53:00 how the Stria example is sequenced and what the parameters mean
56:30 amplitudes in this patch are in decibels (old practice)
57:30 index of modulation units
59:30 beating modulator oscillators
1:01:30 carrier and modulation frequencies more general definition
1:04:00 layout of files in the Chowning study patch (includes a library)
1:06:00 the qlist ("cue list" or "score")
1:08:00 events in the qlist
1:10:00 cross-fade in event 2

msp.ucsd.edu/media/271b.21s/271b-2a-apr7.mp4: Meeting 2a, April 7, 2021
0:00 How to build a piece out of the null piece 
1:30 the patches 0.pd and 1.pd (without or with a qlist stepper/sequencer)
2:30 null piece now can read input soundfiles
4:00 copying the null piece to create a new piece named "aardvark"
4:40 difference between library and score directories
6:00 controls on main window, guts hidden
7:00 keep heavy drawing tasks in subpatches you can hide to save CPU time
9:00 put together a phase modulation sound
10:00 input, DSP, and output windows are connected to control execution order 
11:00 "input1" send/receive and "output1" throw/catch
15:00 modulating oscillator
16:00 carrier oscillator split into phasor~ and cos~
17:00 controlling the two amplitudes
19:00 index of modulation is amplitude of modulating oscillator
20:00 testing the FM pair
23:00 controlling and naming the parameters with "genctl" abstraction
24:00 receive, number box, and genctl working together
25:30 sending a message to set a parameter as in a qlist
27:00 importance of choosing mousable, readable scales (units) for parameters
30:00 fourth-power curve for controlling gains
31:00 the amp4~ abstraction for controlling amplitudes or gains
34:00 controlling the index with amp4~ (later changed)
35:00 first cut at fully controllable FM pair
36:00 "stop" button resets everything (the genctl abstraction does this)
37:30 usnig MIDI units for the two frequencies
38:00 hang a number box off a control object for debugging
39:30 oops, index should use linear units, not fourth-power
40:00 fixing index to be linearly rampable
41:30 amp4~ handles ramping
43:00 genctl makes the control display progress of ramp
43:45 pitfall: sending a pair of numbers to a receiver that didn't expect it
44:30 npack to split elements of the pair
45:20 need an audio rate ramp (line~) to control index to avoid zipper noise
46:30 how to make a default ramp time
51:00 rampable frequencies (ramp in pitch, not frequency0
53:00 wrongness of linearly ramped frequencies
54:00 5-msec default frequency sweep time (vs. 20 for amplitudes)
57:30 pitch sweep working
1:00:00 testing finished FM pair
1:01:00 alternative definition of parameters to specify frequency ratio
1:03:00 controlling the FM pair from main patch
1:04:00 adding a reverberator (rev3~)
1:06:30 rev3~ parameters don't ramp (use an external control line of you want)
1:10:00 reverberator working
1:10:30 using "grab" feature to get parameter snapshot

msp.ucsd.edu/media/271b.21s/271b-2b-apr7.mp4: Meeting 2b, April 7, 2021

1:00 Jonathan Harvey, Mortuos Plango, Vivos Voco.  Quick look at Harvey's paper in CMJ
3:45 difficulty of wielding additive synthesis in general
5:00 generating frequencies in MPVV, partials of Winchester cathedral  bell
7:30 orchestration possibilities of voices plus additive synthesis
9:00 generatong series of MPVV
11:30 sections of piece labeled by generating series
12:30 the glissandi
14:00 stable clusters at beginning and end of glisses suggest pedal tomes
16:00 Rand's technique similar but harmonic series instead of bell
17:00 piece annnounces its series at 1:00
19:00 section 2 announced by its characteristic pitch
19:45 glissandi ni the piece (distorted)
20:30 again, better (not perfect yet)
21:30 sung chord at end
24:30 about the study patch
26:45 opening the study patch
28:00 soundfile chooser to play soundfiles through patch
30:00 the patch has a built-in bell loop
31:00 the sigmund object for sinusoidal analysis
32:00 printed out peaks
33:00 oscillator bank
35:00 comparing analyzed pitches to series in CMJ paper
37:00 separate window for culling and sorting sinusoidal components
40;00 different anayses of sound are similar but distinct
42:00 bell frequencies might not really be stable (Doppler shift)
43:00 research question: what makes some spectra sweet and others ugly?
47:00 glissandi
48:00 making message boxes containing all the components
53:00 sorting components by frequency
55:00 glissing between sorted list of compinents
55:30 selecting components within a pitch range
56:00 verifying vocal chord at end of piece is taken from series
1:00:00 additive synthesis patches always enforce some writing style
1:02:00 why Harvey's piece sounds better than, say Risset's Inharmoniques
1:03:00 Latin inscription on the bell suggests the ethos of the piece

msp.ucsd.edu/media/271b.21s/271b-03a-apr14.mp4: Meeting 3a, April 14, 2021

0:00 looking at recorded sound based techniques earlier than planned
2:00 two currents in electronic music typified by GRM and IRCAM
3:30 studio technique isn't really the opposite of real-time
4:00 study piece is James Tenney, Collage #1
7:00 instantly recognizable sound to a 1963 audience emerges in middle
8:30 study patch is tenney-collage (tenney-collage.zip on PDRP site)
9:00 no patch could do everything interesting; this one just shows some ideas.
11:45 sample rate of patch migth be different from that of recording
13:00 reading and writing soundfiles through PDRP patches
14:00 the sampler itself is in DSP subpatch usnig "clone" object
14:30 messages to "samp" play one "note" using the sampler
15:30 arguments to samp: pitch, gain, duration, sample#, onset, rise, decay
16:30 can play overlapping notes
17:30 microtonal pitches
18:00 making a cluser, several "notes" at same start time
20:00 units of gain (50 for unit gain) and pitch (MIDI, 60 for original speed)
21:00 duration in milliseconds is output duration, not duration in file
22:00 stacking 4 octaves
22:30 need for automation: potentially 1000s of individual notes
24:00 example of audtomation: phased loops
26:30 you might not want to use this in exactly this form.
27:00 period and length in the looper
30:30 timing and emveloping of sampler playback in the patch (diagram)
31:00 onset is where in recording to start
31:30 the amplitude envelope shape
33:45 output is product of envelope generator output and playback signal
34:00 rise time, "duration", and "decay" time (nomenclature is slippery)
35:30 overall length is "duration" plus "decay" time (as in a keyboard synth)
38:00 special case, duration equal to rise time
40:00 setting all three time values equal gives time-symmetric envelope
40:30 sometimes called a Hann window
42:00 why aren't rise, duration, and decay just the same parameter
43:30 you can record audio input live and sample from it
45:00 simple example of automated granular sampling
46:45 changing parameters in a loop
47:45 overlap depends on total playing time and frequency of the loop
49:00 zero rise time to find an attack in the recording
50:00 this can make clicks but we're getting away with it here
50:45 slow rise and fast decay, and vice versa
51:00 desirability of having a choice of decay shapes for more natural decays
53:00 idea: name starting locations instead of giving them numerically
55:00 digression: tanh as a saturation function, different from envelope shape
57:00 digression: using audacity to label onsets and "text" object to store them
1:00:00 another thing you can do with the patch: granular sampling 
1:01:00 exactly periodic repetition gives a tone
1:03:00 getting rid of the periodicity randomizing grain time intervals
1:05:00 diagram of periodic and aperiodic repetition of grains
1:07:00 getting random numbers 0 to 10 in increments of 1/100
1:09:00 fixed "samp" messages in message boxes aren't very flexible
1:09:30 the looper example demonstrates parametrized messages
1:10;00 should add time reversal to dolist (patch doesn't do it yet)
1:11:00 item in dolist: you can't track down a speaking "voice" to alter it
1:14:00 more dolist: this patch only does stereo in and stereo out
1:15:00 more dolist: separate streams ("tracks") of samples
1:16:45 more dolist: multichannel output and spatialization

msp.ucsd.edu/media/271b.21s/271b-03b-apr14.mp4: Meeting 3b, April 14, 2021

0:00 aleatoric processes
4:30 making duration variable
6:00 using random time both for metronome and for duration of sample
7:00 changing playback location in time (onset)
7:30 range from 500 to 1500 (1499 to be exact)
9:00 saple rate correction seems not to be working yet
11:00 more than one dollar substitution in message boxes: "pack" object
13:00 "float" object stores first value so that it doesn't trigger "pack"
13:45 "bang" sets the pair of numbers off, and you can asynchronously set them
14:30 you can do the opposite (make "pack" output a message on either input)
17:00 random number fo r onset, range chosen to fall within guitar solo
17:45 using trigger to put number in non-first inlet of pack and trigger it
19:30 all these parameters probably need names.  Name ranges in pairs
20:00 random-onset-start and random-onset-range variables and number boxes
23:00 receives can go into control (message) version of line
24:30 default time grain of 20 milliseconds
27:00 control names for on/off and for time duration
28:00 suggestion for messages that can't ramp: use "f" to be sure of message
30:00 putting an event in a message box to start randomized-playback process
32:00 adding pitch variation
34:30 three elements to pack together now, ordered by trigger "b f b"
35:30 adding randomizer for pitch
37:00 grain of randomization should be much smaller than 1 (1/1000 here)
38:30 you might want to specify middle of random range instead of start
41:00 trying to imitate beginning of Tenney piece
44:00 how could you make this loop?
45:30 you can have multiple processes running at once
47:00 another automated patch: granular sampler
48:45 variables we'll want: pitch, timing, duration, onset
50:00 bad but effective way to replace a name systematically (random to grainer)
53:00 inter-onset interval (time delay) no longer same parameter as duration
55:00 combine pitch, onset, and duration into a "samp" message
59:00 granular synthesis result from vocal portion of recording
1:00:00 randomizing pitch variation
1:01:00 randomizing onset makes a more diffuse sound
1:02:00 oops, another process had been running, rerunning the tests
1:06:00 possibility: sampling live instruments, playing patch as an instrument
1:07:00 more possibilities: attach MIDI controllers and/or use presets
1:09:00 inside the cloned "stereo-sample-reader" patch
1:10:00 inlet for messages to start playback, issued by "clone" object
1:11:00 more capabilities of "clone"
1:13:00 how you could adapt this patch to your own ends
1:14:00 Polansky has a paper about collage #1 on his website

msp.ucsd.edu/media/271b.21s/271b-04a-apr21.mp4: Meeting 4a, April 21, 2021

0:00 Hidegard Westerkamp's Beneath the forest floor
0:30 musique concrete and soundscape composition compared
3:00 writings and videos available about the piece 
3:30 Inside Computer Music by Clarke, Dufeu, and Manning
6:00 downloadable materials on their website including apps to explore pieces
7:00 Westerkamp's videos describing the genesis of the piece
14:30 why one recording of two ravens didn't work i nteh piece
16:00 another raven recording that proved useful
19:00 the raven recording in audacity
20:00 1/70 - second-long ululations
21:00 TaCEM app shows the layout of the entire piece
24:00 reconstruction of piece in graphical form showing recordings and processes
25:20 slowed-down raven sound in TaCEM app
27:00 reverberance extended by slowing the recording down
29:00 trying it in audacity
32:00 three separate calls used so that it doesn't sound like sampling
34:30 overall view of piece in audacity.
35:00 Raven calls dominate forst part of piece
38:00 sounds of wind, lifeforms, and water
39:00 the sound of the motorized saw answers teh raven
40:30 quiet section that recalls beginning
41:00 pitched sounds emerge in second half
44:00 other bird sounds are sources of pitched sounds
44:30 the wren recording is high-pass filtered and slowed down
45:00 7 kHz salient frequency
47:30 time stretching (just an example, not the original technique)
48:00 more faithful imitation using reverberation
49:30 related sound from in piece (but (oops) a different bird)
50:00 hovering chords not unusual in traditional tape music
51:00 treatment of rusing water with delays
53:00 spatializing stereo recordings
54:00 close microphone pair gives impression of stereo field
56:00 Westerkamp: not sampling and manipulation, but recording and processing
57:30 melodic material from thrush song
1:03:00 1990-era technology
1:04:00 pitch fields in recording-based music

msp.ucsd.edu/media/271b.21s/271b-04b-apr21.mp4: Meeting 4b, April 21, 2021

0:00 adding to Tenney patch to imitate techniques in Westerkamp
1:30 need to filter birdsong to get clean sounds
3:00 closer look at thrush sound in audacity
5:00 start times and durations of three thrush notes
7:00 playing thrush notes in Tenney patch
10:00 wrong transposition (fixed after class)
14:00 back to filtering
18:00 butterworth low-pass and high-pass filters
19:30 testing filters with noise input
25:00 more about filters
26:30 improving on simple band-pass filter by pairing two detuned ones
31:00 frequency response of Butterworth versus simple low/high-pass filters
32:00 Butterworth filter design
36:00 three thrush notes into sampler
37:30 reverberation using rev3~ object
44:00 testing reverberator using an impulse as nput
48:00 sampler into reverberator
52:00 fix filter cutoff to match transposition
55:00 why ramp up and down over 100 msec - gain ramp as modulation
57:30 frequency response of a ramp
58:30 the faster the envelope, the higher-frequency the aliasing
1:02:00 these ramps can be slow but in other cases you want a fast one
1:02:30 to preserve the attack in a recording make rise time instantaneous
1:04:30 gating (applying a noise gate to a recording)

msp.ucsd.edu/media/271b.21s/271b-05a-apr28.mp4: Meeting 5a, April 28, 2021

0:00 Trevor Wishart's Imago: wavesets and phase vocoders
2:00 the acousmatic musical style
2:30 same techniques as in soundscape composition but different musical style
4:00 idea of deriving a piece out of a small sonic seed: Xenakis's P/H Concrete
6:30 seed sound for Wishart: two whiskey glasses clinking together
9:00 formal approach: form imposed by cmoposer, not emerging from soundscape
10:30 characteristic acousmatic move: gathering energy and release
12:30 close look at the seed sound using phase vocoder
17:00 time-frozen decay of the sound
19:00 uses of phase vocoder in Imago
19:30 frequency shifting (might not actually be the thing found in the piece)
23:00 pitch shifting and frequency shifting contrasted theoretically
28:00 incitement: use both pitch and frequency shifting to squash a spectrum
31:30 frequency shifting (single sideband modulation) compared with ring mod
35:00 more about phase vocoder
37:30 location in source file (center of analysis window)
38:00 window size
39:00 Hann window (same envelope as in granular example from Tenney)
41:00 how the Hann window modulates the sound
43:00 spectral stamp imposed by Hann window is 4x findamental frequency wide
44:00 analyzing a sum of sinusoids: frequency resolution
45:00 2048 points (fundamental is about 24 Hz.) - 100 Hz. resolution
47:00 100 Hz. is about low A flat.
49:00 testing phase vocoder on a combination of sinusoids
54:00 what happens when the window is too small (i.e., sinusoids too close)
56:30 fundamental tradoff: time resolution versus frequency resolution
1:00:00 two frequencies in seed sound, 1763 and 1863
1:04:00 time stretching using the phase vocoder patch
1:06:00 why the phase vocoder sounds spacy (or phase-y)
1:07:00 phase locking.  Frequency domain is separated into "bins"
1:13:00 phase vocoder on speech
1:16:00 speech stretched usnig very short analysis window

msp.ucsd.edu/media/271b.21s/271b-05b-apr28.mp4: Meeting 5b, April 28, 2021

0:00 more about phase vocoder: compared with reverberation
8:00 vibrato via pitch shifters
9:00 2700-ish frequencies can be made to sound like a vocal formant
11:00 waveset manipulation (specifically waveset duplication)
18:00 Two Women, Four Voices: simpler examples of Wishart's techniques
19:00 part 3 (Ian Paisley): phase vocoder.  Vibrato and also too-small windowing
20:00 part 2: waveset duplication on Diana's voice
24:00 waveset duplication tested on two sinusoids
26:00 controls: minimum number of repetitions and minimum waveset length
34:00 high frequencies can re-emerge as formants
36:00 waveset duplication tested on Diana's voice
36:30 waveset duplication compared to wavetable oscillation
39:00 other waveset operations (not ready in this patch)
40:00 waveset substitution
42:00 waveset averaging
44:30 waveset duplication on seed sound
48:00 time-stretched seed sound through waveset duplication
52:00 why waveset duplication sometimes puts out higher pitches
58:00 Wishart's emphasis on hearing the source sound rather than the technique
1:02:00 even though the form is classical, result is guided by what he finds
1:05:00 Composer's Desktop Project (CDP), well-established software
1:06:00 using phase vocoder to shape envelope

msp.ucsd.edu/media/271b.21s/271b-06a-may5.mp4: Meeting 6a, May 5, 2021

0:00 Natasha Barrett, Little Animals and (mostly) The Lock, from Hidden Values
3:00 debts to Dennis Smalley, Jonty Harrison, (early) Trevor Wishart
8:00 "wav" and "aiff" can't hold long soundfiles with many channels
10:00 classig GRM-ish sounds in Little Animals
11:30 sound projection - technique/style in which a stereo tape is spatialized live
13:30 realistic-sounding, "present" images in stereo recordings
16:00 foreground and background layers
17:00 resonant filterbanks.  Another example from Noanoa by Saariaho
19:00 The generating flute multiphonic
21:30 tradition (or trope) of discursive sounds against a static pitch field
24:00 The Lock avoids many electroacoustic-music tropes and keeps others
24:30 pitch fields in The Lock made by time-stretching and filtering the voice
27:00 spatialization is central to Hidden Values
28:30 spatial paths in concert with musical gestures and timbral evolutions
31:00 Barrett shows her setup in video from TaCEM
32:30 IRCAM spatializateur ("spat") 2 views of spatial paths
34:00 Nuendo DAW feeding 32(?) channels into IRCAM spat via jack
36:00 acousmatique-style spatial projection is not same as cinematic spatialization
36:30 cinematic-style hearkens back to Stockhausen in the 1950s
37:30 angle cues via panning and distance cues via filtering and reverberation
38:30 John Chowning's paper The Simulation of Moving Sound Sources
40:00 three approaches to cinematic spatialization: Ambisonics, VBAP, and WFS
42:00 VBAP (Vector based amplitude panning)
43:00 VBAP simulates direction but not distance of a virtual source
45:00 cinematic spatialization as used to simulate moving sources
46:30 Gestalt "common fate" grouping to fuse different source sounds
47:00 this is described in Four Criteria of Electronic Music by Stockhausen
49:00 Barrett placing three sounds spread out in space but moving together
50:30 simulating distance on top of VBAP
53:00 Ambisonics as a way to represent a local sound field (not cinematic)
54:30 to fully represent the sound field in a room takes about a million speakers
57:00 a soundfield is a superposition of plane waves in all directions
58:00 at a single point there are only 4 independent channels
58:30 but humans pick up more about the sound filed than from a single point
59:30 two-dimensional ambisonics as a function of time for every direction
1:01:00 if you sample the possible directions you get a Nyquist-like cutoff
1:02:00 to first order: DC and first-harmonic sinusoid around a circle
1:03:00 "radiation patterns" for ambisonics: omni and figure-eights
1:04:00 weighted sums of three basis patterns to make others such as cardioid
1:05:00 you can theoretically record first-order sound field at a single point
1:06:00 recording and playing back first-order Ambisonic signals
1:08:00 Ambisonic channels are in (or 180 degrees out of) phase (no delays)
1:10:00 issues of polarity and diffuse-ness in Ambisonic reproduction
1:12:00 higher-order Ambisonics.  Second-order requires 5 channels
1:13:00 equivalent to sampling the circle of directions into 5 points
1:14:00 linearity of Ambisonics allows superposition of sound fields

msp.ucsd.edu/media/271b.21s/271b-06b-may5.mp4: Meeting 6b, May 5, 2021

0:00 Wave field synthesis (WFS)
2:30 imagining a faraway sound source heard through a series of open windows
3:30 special case: all sources ("windows") get the same signal
5:00 beam width depends on wavelength of sound and total width of source array
6:00 side-beams (additional directions of radiation)
8:00 spacing of speakers matters and should be a wavelength or less
9:00 aliasing in the angle of radiation depends on individual speaker spacing
10:30 projecting at an angle requires that you delay the signals in the speakers
13:00 requirement to avoid spatial aliasing
14:30 true WFS would require a speaker every 1/2 inch
16:00 WFS can theoretically simulate a sound source inside the listening area
17:00 VBAP and Ambisonics can only simulate faraway sources
18:00 one-dimensional wavefield arrays actually output cylindrical waves
19:00 OK, you could use a pizza-box shaped listening space
21:00 problems that arise when audience is dispersed in a listening space
23:00 problems when panning to simulate virtual sources between speakers
25:00 yet another problem: Hass (precedence) affect
29:00 in extreme case, one foot off-axis can cause Hass trouble
31:00 Ambisonics seems to get a larger listening sweet spot than VBAP
32:00 but directionality is more diffuse (less clarity of image)
34:00 highly present sounds of traditional acousmatic music benefits from "projection"
35:00 summary: advantages and disadvantages of VBAP, WFS, and Ambisonics
39:00 implications for listening to music in listening spaces in the future
40:00 binaural reductions of loudspeaker spatialization
41:00 how binaural spatialization works
43:00 head-related impulse responses and transfer functions (HRTFs)
45:00 spatial perception requires freely moving head
47:00 the ORTF mic setup
48:00 mixing an ORTF pair to a monaural signal
50:30 why worry about cinematic spatialization at all?
51:00 example: Janet Cardiff's 40-voice motet
52:00 octagonal loudspeaker arrays frequently used in electronic music concerts
53:30 affordance: clarity from separating "voices", moving or not
55:00 concert halls blur out details of amplified sounds
58:00 music in front of listener versus music played from behind
59:00 suggestion: put speakers and live musicians on stage together
1:02:00 more on Barrett, The Lock - the nature of the sound sources
1:02:30 TaCEM screen showing geneologies of materiels
1:04:00 processing is mostly time stretching, filtering, and reverberation
1:06:00 electroacoustice-style pitch fields easy to make beacuse of source
1:08:00 spatial separation to enhance clarity
1:09:00 apparently used all three spatialization models for IRCAM performance
1:10:00 Spatialisateur used to simulate binaural recording of loudspeakers via HRTFs
1:11:00 advantages: no room reverberation, and listener is in sweet spot
1:13:00 simple, arcing spatial paths
1:16:00 binaural spatialization better at sides than front

msp.ucsd.edu/media/271b.21s/271b-07a-may12.mp4: Meeting 76a, May 12, 2021

0:00 Laurie Spiegel, Appalachian Grove
1:00 Groove system at Bell Labs - computer controlled analog electronics
2:00 this was Mathews's first foray into real-time computer music
3:00 genesis and design of "hybrid" studios
5:00 related to sequential drum (Mathews and Moore)
6:00 what the piece sounds like and how it was generated
8:00 control inputs to analog patch - ADSR triggers, pitches, filter parameters
12:00 center-panned voice only speaks on on-eighth-notes
13:00 choice of pitches
14:30 the ADSR envelope generators
16:00 what happens when sustain level changes while ADSR is in D or S phase
18:00 can enter release phase while in A or D
20:30 study patch: signal paths first, then control
23:00 ADSR abstraction
24:00 output is line segments so usually need to apply a nonlinear function
24:30 band-limited sawtooth generator
25:30 resonant filter
26:00 panner
27:00 control panel: tempo in attacks per minute
28:00 ADSR abstraction has a peak-amplitude parameter (unlike analog ones)
30:00 preview of mechanism to build control panel
31:00 six parameters for each ADSR (one for amplitude and one for VCF)
32:00 pitch, pan, and q are per-voice (three each eventually)
32:30 units for ADSR envelopes: times in milliseconds
34:30 "peak" is in whatever units used by what you send the ADSR to
35:30 sustain level is a percentage of peak
36:30 effect of changing duration of DASR triggering pulse
39:30 rearranging the patch to contain three oscillators
42:00 control (message) line objects to allow ramps in ADSR parameters
43:00 using the new parameters
44:00 testing ramping the ADSR release time
45:00 comparison of LFO versus ramps to control parameters
47:00 aliasing if paramter is changing faster than ADSR is triggered
50:00 historical hardware was (sort of) using audio-rate parameter changes
52:00 pitch is fixed for duration of each individual note (not continuous)
54:00 abstraction for the oscillator so we can make 3 copies
56:00 pan controlled inside abstraction
57:00 individualize pan using a control with a per-copy name
58:00 gotcha: don't retype an abstraction if you're editing the interior
59:00 use outlets for output signals and add them outside the abstraction
1:00:00 parameters for filter: center frequency (as pitch) and q
1:02:00 need two filters, one for each output channel
1:05:00 managing triggers for ADSRs (delay equal to note duration)
1:06:00 random choice of which voice to trigger each tick
1:07:00 pitch1, etc, parameters to set pitch
1:07:30 rebuilding control panel
1:09:00 genctl takes an argument to specify default (reset) value
1:10:00 testing the three-voice patch driven by a metronome
1:11:00 imitate detuning of analog oscillators
1:12:00 grab all the parameters and put in a message box
1:14:00 rhythmic detail in the piece: center voice is accented
1:16:00 center-panned voice only appears on on-sixteenths
1:17:00 no hardware sequencer (it was in the computer program)
1:18:00 computer also had live analog inputs

msp.ucsd.edu/media/271b.21s/271b-07b-may12.mp4: Meeting 7b, May 12, 2021

0:00 photo of control panel for (I think) GROOVE system
1:30 center frequency, bandwidth and q of a resonant filter
8:30 Moog ladder filter compared to vcf~ object
10:00 affordances of computer-controlled analog versus digital synthesis
12:30 imitating rhythmic pattern in Appalachian Grove
22:00 will skip doing the accents correctly
23:31 control of pitches
26:00 the value object, used to read next-pitch into correct voice
27:00 why you don't need a trigger to control order of pitch versus ADSR
31:30 selecting pitches from a collection
33:00 one way to make a pittch set: 7-element array with a message to fill it
34:30 perhaps better alternative: use a text object to hold the pitch set
38:00 array for pitch probabilities
41:00 bob~ object: ladder filter, alternative to vcf~
53:00 adding reverberation
54:00 automatic control panel generator
55:30 looking at saved patch to learn how to create objects
57:00 windows in Pd can be sent messages by window name
1:00:00 make-ctls sub-patch.  Catching the first in a sequence of messages
1:02:00 counter to set y location of controls in new panel
1:03:00 supplying default arguments to messages using unpack
1:08:00 time stretching (sort of) using long reverberation
1:11:00 short versus lnog sinusoidal burst into infinite reverb
1:16:00 this is probably what was done with bird sounds in Westercamp piece
1:16:30 other ways: phase vocoder (e.g., Wishart); resonant filters (Saariaho)

msp.ucsd.edu/media/271b.21s/271b-08a-may19.mp4: Meeting 8a, May 19, 2021

0:00 invited talk: Kerry Hagan (Univ. of Limerick), ...of pulses and times...
4:30 generative music and electronic music genres
6:00 inspired by Nick Collins.  Probabilistic analysis of beat music
7:30 algorithmic versus generative music and electronic music genres
8:30 another inspiration: Mortom Feldman's metric modulations
9:30 layering tempos with different pulses
11:00 playing the piece from the patch (note: recording is only one channel)
18:00 why not use samples
19:00 top-down design of piece
20:00 score is series of messages
21:30 metric modulation subpatch
24:00 metronome objects generating different subdivisions of beat
25:00 non-unifirm probabilities
27:00 10 percent nonuplets, etc.
28:00 random selection of sounds depending on beat
30:00 occasional detour nito triplets (dubstep reference)
33:00 sound generator subpatches
35:30 usnig vline object to make stuttered pulses
37:30 attacks don't exactly repeat so that successive events aren't identical
39:30 FM instrument borrowed from another piece, Morphons and Bions
41:30 chirps, drones, and swoops
43:00 patch can play automatically or be performed live
45:00 automatic meter changes
47:00 Pd programming style
48:30 preparing patches to be played without composer present
50:00 control surfaces
52:00 memoryless processes and longer time durations
54:30 piece can sound different every time although the form is fixed
57:00 comparison with fixed-media pieces
1:00:00 real time is fundamentally different compositionally
1:01:00 multichannel sound and real-time spatialization
1:03:00 no filtering used (but some broad-band sounds and some narrow ones)
1:06:00 using filters in generatove or algorithmic music
1:06:30 serialism and representation of all musical values as numbers
1:08:00 role of accident in stochastic music
1:11:00 narrative and musical form
1:14:30 how strictly should a performance follow the "score" (message sequence)

msp.ucsd.edu/media/271b.21s/271b-09a-may26.mp4: Meeting 9a, May 26, 2021

0:00 more on Manoury, Pluton.  Running patch using recorded piano and MIDI
6:30 "section 100" - qlist1.txt and score1.txt
7:30 score follower compares incoming MIDI from piano with score
8:00 example: C-D-E-F-G-F-E-D-C-C-C-C-C-C in score and repeated Cs from keyboard
9:00 jumps to event 19
10:30 concert setup for Pluton.  MIDI interface, PianoBar designed by Buchla
11:00 audio conections
11:30 logical diagram: frequency&pitch shifters, reverb, spatialization
12:30 oscillator bank controlled by FFT analysis of piano; and live sampler
13:30 how MIDI input is used: first, velocity-controlled spatialization
14:30 second use for MIDI input: training a Markov chain
15:30 third use: score following.  Barry Vercoe and Roger Dannenberg
16:30 score-following driven sequencing (qlist)
18:30 don't need to send MIDI back to "play" piano
19:30 score following algorithm.  Possibility of very tight synchrony
21:30 explanation of how the example from 8:00 worked
24:30 finding best possible path connecting score with performance
27:00 each path has a conditional probability (incorrectly called "likelihood")
30:30 most probable path implies where we want the computer to be
32:00 equivalent to Viterbi algorithm for a particular hidden Markov chain
34:00 related term: "dynamic programming"
35:30 limitation: this approach doesn't take advantage of timing of events
36:00 why people don't often use score following today
37:30 polyphonic score following
39:00 only using note onsets as input to score following is a severe limitation
44:00 research problem: score following for Pluton using audio alone
46:00 "notes.txt" file in patch show what labels are where in score
47:00 Event 31 (section IV): FFT-controlled oscillator bank
47:30 section II introduces oscillators and Markov chain and alternates them
48:30 section III Markov alone
50:30 "gumbo" objects that control pitches of oscillators
51:00 oscillator control panel.  Capital and small "o" .  Oto2, Oto4, etc.
56:00 amplitudes of oscillators controled by FFT analysis as in a vocoder
58:00 spectral shift and response speed controls
59:00 why this doesn't sound like a vocoder used in the usual way
1:01:00 usnig the four 12-voice banks as one larger bank (gumbo5 object)
1:02:30 why the "gumbo" object has setparate "set" and "bang" messages
1:03:30 "mode 1" loads pitches into the oscillators from piano keyboard
1:08:00 aliasing effect in FFT analysis if speed set high
1:10:00 setting speed to zero freezes spectrum
1:11:00 cross-fading between spectra
1:12:00 different wavetables
1:14:00 patch to simulate piano (audio and MIDI).  Section III, events 1-6
1:18:00 Pd patch adapted from original Max patch

msp.ucsd.edu/media/271b.21s/271b-09b-may26.mp4: Meeting 9b, May 26, 2021

0:00 Markov chain object ("mark" receive name)
1:00 recording MIDI into the Markov chain which controls a sampler
1:30 "trec" message records; "play1" plays back
3:00 Record a chain C-D-E-F-G-C
3:30 Oops, need to clear the chain first
4:00 ambitus control
5:00 transition probabilties in Markov chain taken from MIDI velocity
6:00 time limit between notes to make a transition
6:30 restart parameter in case chain gets stuck
8:00 common pitches between different pitch sets to make transitions
9:30 how rhythm is input from MIDI
11:00 if Markov chain has absorbing states it lasts a random amount of time
13:00 section 2: oscillators first, then Markov chain at event 7
16:00 Markov chains have faded out; oscillators take over
18:00 reemergence of Markov chain
18:30 recording arpeggios into sampler
19:00 ambitus is at zero
20:00 Markov chain disappears a second time
21:30 playing chords into Markov chain (not demonstrated)
23:30 imperfect separation between interaction and score (events 43:1-5)
26:00 three metronomes at different speeds
27:30 event 3 records from piano to sampler bank 2
30:00 pitch shifter ("harmonizer") to make celeste effect
32:30 playing through section V, part E
38:30 granular time stretching in part F
41:00 time stretching played without piano part
42:00 sample played without time stretch
43:00 playback at different speeds. The "rspeed" control
46:00 events 44:1-3 (section E part F)
49:00 ending of piece, 64-to-one time stretch
50:00 spatializer, repeated patterns with variable speed
51:00 tables (arrays) giving x and y coordinates for spatialization
55:00 velocity control of spatialization
57:30 velocity controled crossfades between oscillator banks
1:00:00 velocity control conects piano to electronics
1:05:00 research problem: make a cheap, robust piano MIDI detector
1:06:00 designing pickups for individual piano strings
1:08:00 a more recent algorithm (from Manoury, en Echo): "3F"
1:11:00 three generating pitches generate inharmonic spectra
1:13:00 secret sauce: linear integer combinations of generating frequencies
1:15:00 example: using incoming pitch as one of the three frequencies



-----------------------------------------------------
Music 270b. Analysis of musical sound.
Spring quarter 2022.
course materials: msp.ucsd.edu/syllabi/270b.22s/
recordings: msp.ucsd.edu/media/270b.22s/
contents of recordings:

msp.ucsd.edu/media/270b.22s/270b-01-mar28.mp4: Meeting 1, March 28, 2022

0:00 This course is about signal analysis, not cognitive science as advertised
5:00 about the sonic challenges
6:00 first sonic challenge: the call of the white bellbird
21:30 short-time Fourier analysis (STFT)
26:00 tools used in this course: Pd, Audacity, Julia
29:00 storing a sinusoid in an array in Pd
33:30 classical envelope follower: rectify and low-pass filter
36:30 phase of a sinusoid
40:00 phase quadrature
42:00 complex sinusoid as real sinusoid pair
44:30 magnitude of a complex sinusoid
47:30 the hilbert~ object
50:00 measuring the magnitude of a real sinusoid using hilbert~
54:00 sinusoids in nature: time-varying frequency and magnitude
56:00 ring (amplitude) modulation to change frequency of a sinusoid
58:30 aliasing a frequency to almost zero
59:30 using phase quadrature to modulate an oscillator
1:01:30 low-pass filtering the modulated sinusoid
1:02:00 measuring the magnitude of the sinusoid
1:04:00 reaction time of magnitude measurement
1:05:00 trade-off, stability versus reaction speed


msp.ucsd.edu/media/270b.22s/270b-02-mar30.mp4: Meeting 2, March 30, 2022

1:00 finding and loading the example patches from classes.  Some Pd basics.
5:30 complex-mpy~ abstraction
7:00 ampliude detection from previous class understood as a frequency shift
14:30 real and complex sinusoids.  Real sinusoid is sum of two complex ones
21:00 measuring amplitude and phase of a complex sinusoid
27:00 single sideband modulation as product of complex sinusoids
32:00 x/y plot using Pd of a complex sinusoid.
33:00 Positive frequency sinusoids travel counter-clockwise in complex plane
40:00 Measuring phase differences between neighboring sinusoids
42:30 complex conjugate negates frequencies of sinusoids
43:30 phase difference3 between an incoming sinusoid and a known sinusoid
51:00 analysing complex spectra (tones with more than one partial)
56:00 spectrum of frequency-shifted complex sinusoid making 6 components visible
1:01:00 low-pass filtering a sum of complex sinusoids
1:03:00 Fourier analysis of a periodic tone
1:08:00 Short-time Fourier transform
1:09:00 Window as a segment in time
1:13:00 windowing a non-periodic signal
1:15:00 trade-off between time and frequency resolution


msp.ucsd.edu/media/270b.22s/270b-03-apr04.mp4: Meeting 3, April 4, 2022

0:00 terminology: DFT, FFT, STFT (short-time FT), WSTFT (windowed STFT)
4:00 windowing and window functions
6:00 magnitude spectrum of a sawtooth; discontinuities in waveforms
8:00 analyzing a sinusoid not tuned to window size
10:30 boxcar (rectangular) window function
12:00 continuous-time Fourier transform of a windowed sinusoid
13:00 total energy of a continuous-time signal
14:30 Hann window function in continuous time
21:30 Hann window in discrete time (patch example 5.hann-window.pd)
29:00 analyzing and playing the contents of a window as a periodic signal
39:00 windowed analysis of a constant signal. Main lobe and sidelobes
44:00 spectrum of a square wave
50:00 spectrum of the Hann window function and a Hann windowed sinusoid
57:00 derived boxcar window spectrum
1:03:00 sinx function
1:06:00 measuring the spectrum of a short square pulse (boxcar window function)
1:10:00 derived spectrum of Hann window
1:15:00 spectrum of Hann windowed sinusoid
1:17:00 superposition of two sinusoids


msp.ucsd.edu/media/270b.22s/270b-04-apr06.mp4: Meeting 4, April 6, 2022

0:00 How the discrete Fourier transform (DFT) works
3:00 fundamental frequency is sample rate divided by N
10:30 complex-valued Fourier transform of a constant signal
12:00 formula for discrete-time Fourier transform (generalization of DFT)
20:00 normalizing the DFT
21:30 inverse DFT to recover original array
27:00 DFT as way to separate frequency ranges in a signal
32:00 freqency bins and Fourier analysis between bins
46:00 relationship between sinx function and Fourier transform of a sinusoid
49:30 Pd example patch: 64-point DFT of a sinusoid
56:00 changing the starting phase of the sinusoid
1:00:00 graph of analyzed waveform
1:09:00 how the FFT works
1:16:00 DFT as multiplication by a unitary matrix


msp.ucsd.edu/media/270b.22s/270b-05-apr11.mp4: Meeting 5, April 11, 2022

1:30 Portnoff paper on WSTFTs; channels of transform are low-bandwidth signals
5:00 tail behavior of Hann window spectrum
9:30 Blackman-Harris window
15:00 zero-frequency bin as low-pass filter
20:00 output can be downsampled to a hop size of N/4
24:00 situation for bin 1
31:00 signal is multiplied by Hann window times complex exponential
34:30 alternating signs of consecutive bins when analyzing a sinuaoid
39:00 measured phase precesses for nonzero bins
48:00 analysis of sinusoid tuned to first bin, H=N/4
53:00 measuring frequency of a sinusoid using phase precession (phase vocoder)
54:00 multiple possible vaules of frequency separated by 1/H cycles per sample
59:00 aliased frequencies should differ by at least 4 bins so we need H <= N/4
1:05:00 analyzing complex sums of sinusoids
1:10:00 needed separation between sinusoids to resolve them
1:12:00 demonstration using Pd help window I02.Hann.window.pd
1:17:00 N=2038 at 48000: 23 Hz. per bin, 43 millisecond window.


msp.ucsd.edu/media/270b.22s/270b-06-apr13.mp4: Meeting 6, April 13, 2022

0:00 the white bellbird analyzed and resynthesized
9:30 06.pvoc-squeezed-spect-graph.pd in class patches
18:00 peaks in bellbird spectrum
25:00 synthesizing a formant 
29:00 pulse trains by waveshaping sinusoids
35:00 seeing between the bins in a Fourier analysis by zero padding
46:00 thought experiment: pinched windows and signal variations within windows
47:00 applying Hann window a posteriori by convolving window kernel
56:00 Peeters (ICMC 1999) - analyzing a chirp (swept sinusoid)
1:05:00 rate of frequency change adds to measured bandwidth in WSTFT

msp.ucsd.edu/media/270b.22s/270b-07-apr18.mp4: Meeting 7, April 18, 2022

0:00 Fourier transform of a sinusoid as continuous function of frequency
2:00 printing out discrete Fourier transform of a Hann windowed sinusoid
3:30 peak is four bins wide.  Relating this to figure in Techniques chapter 9
5:30 numerically verify that phases of neighboring bins differ by pi radians
11:00 phase vocoder determines frequency by measuring phase change over time
15:00 phase change in a neighboring bin over time should be the same
15:30 setting hop size to one sample as a special case
17:00 how to measure phase precession using a single window
24:00 is a bigger hop size better for accuracy?
28:30 difficulty analysing sweeping frequency
30:00 frequency sweep is second derivative of phase, so numerically fragile
31:00 demonstration using sigmund~ object on the white bellbird recording





[remaining videos not yet indexed.]