Synthesis with Code

Chapter 21. Meeting 21, Languages: Synthesis with Code 21.1. Announcements • Music Technology Case Study Final due Today, 24 November • Sonic System Project Report due Thursday, 3 December • Last quiz: Thursday, 3 December 21.2. Quiz Review • ? 21.3. Modern Music-N • 1985: Csound • Command-line program • Currently available for many platforms, with many interfaces • http://www.csounds.com 21.4. Csound: Score, Orchestra, and CSD Files • Orchestra (a .orc file) defines synthesis processing and interconnections (instruments) • Score (a .sco file) defines control information (events and parameters) • A CSD file combines the score and orchestra into a single file delimited by XML-style markup • Outermost: • Orchestra: • Score: 21.5. Csound: Orchestra Syntax: Instrument Blocks • A quasi programming language, closer in way to an assembly language Orchestra procedures can be extended with Python (Ariza 2008) • Consists of statements, functions, and opcodes • Opcodes are unit generators • Comments: start with a semicolon • Instruments in blocks, named with a number: • Start marker: “instr” and a number • End marker: “endin” • Trivial example instrument definition instr 100 ; comments here! endin 21.6. Csound: Orchestra Syntax: Signals • Signals are created and interconnected (patched) within instrument blocks • Signals: cary streams of amplitude values as numbers within the dynamic range (16 bit audio uses integers from -32768 to 32768) • Signal paths (like patch cords) are named variables • Signal variables can be at different rate resolutions depending on the first letter of the variable name • a-rate: Audio, name starts with an “a” (e.g. aNoise) • k-rate: Control signals, name starts with an “k” (e.g. kEnvl) • i-rate: Initialization values, name starts with an “i” (e.g. iFq) • Example aNoise • random -12000, 12000 Opcodes: unit generators • Syntax uses spaces and commas to delineate: Provide variable, opcode, and space-separated parameter arguments destinationSignal opcodeName arg1, arg2, ... • • Example: “random”; takes two arguments: min and max • Example: “outs”: takes two arguments: two signals Example instr 100 aNoise random outs -12000, 12000 aNoise, aNoise endin 21.7. Csound: Score Syntax • A list of events and parameters given to instruments in the orchestra • Provide a space-separated list of at least three values on each line: i i i instrumentNumber instrumentNumber instrumentNumber startTime startTime startTime duration duration duration p4 p4 p4 p5 p5 p5 ... ... ... • Additional parameters (called p-fields) can be added after duration and provided to the instrument in the score • Example: Two events for instrument 23 lasting two seconds, starting at 0 and 5 seconds i i 23 23 0.0 5.0 2.0 2.0 21.8. Csound: Rendering an Audio File • Call the CSD file with the csound command-line application on the CSD file to render audio csound -d -A noise.csd -o out.aif • • Provide a “flag” to indicate type of audio output • -A (aiff output) • -W (wave output) Give sampling rate, control rate, and number of channels in a header sr kr ksmps nchnls = = = = 44100 4410 10 2 • Example: a noise instrument • Example: tutorial-a-01.csd sr = 44100 ksmps = 10 nchnls = 2 instr 100 aNoise random outs -12000, 12000 aNoise, aNoise endin i 100 0 2 i 100 3 2 i 100 6 2 21.9. Csound: GEN Routines and Wave Tables • Some opcodes require a wave table identification number as an argument • Wave tables are created with GEN routines in the score file, before events are listed • Example: a GEN routine used to create a 16384 point sine wave as a wave table f • 99 0 16384 10 1 Oscillators require a wave table to provide a shape to oscillate The oscili opcode oscillates (and interpolates) any shape given in the f-table argument aSrc oscili amplitude, frequency, functionTable • Example: two instruments, a noise and a sine instrument • Example: tutorial-a-02.csd sr = 44100 ksmps = 10 nchnls = 2 instr 100 aNoise random outs -12000, 12000 aNoise, aNoise oscili outs 12000, 800, 99 aSine, aSine endin instr 101 aSine endin f 99 0 16384 10 1 i i i 100 100 100 0 3 6 2 2 2 i 101 2 6 21.10. Csound: Scaling and Shifting Signals • Example: using a scaled sine wave as an envelope of noise • Assignment (=) and operators (+, *) permit mixing and scaling signals • Example: tutorial-a-03.csd sr = 44100 ksmps = 10 nchnls = 2 instr 102 aEnvl aEnvl oscili .5, 6.85, 99 = aEnvl + .5 aNoise aNoise random -12000, 12000 = aNoise * aEnvl outs aNoise, aNoise endin f 99 0 16384 10 i 102 0 2 i 102 3 2 1 21.11. Csound: Adding Parameters to Score and Orchestra • pN (p1, p2, p3, p4, ...) variables in orchestra permit additional parameter values to be provided from the score to the instrument • Design of instruments in the orchestra requires choosing what parameters are exposed in the score • Example: tutorial-a-04.csd sr = 44100 ksmps = 10 nchnls = 2 instr 102 iDur = p3 iTrem = p4 aEnvl aEnvl oscili .5, iTrem, 99 = aEnvl + .5 aNoise aNoise random -12000, 12000 = aNoise * aEnvl outs aNoise, aNoise endin f 99 0 16384 10 1 i 102 0 2 6.2 ; fourth parameter is frequency of sine envelope i 102 3 2 23 i 102 6 2 45.6 21.12. Csound: Adding Filters • Numerous opcodes exist to explore a wide range of common synthesis tools • Low pass filter aDst lowpass2 aSrc, cutoffFrequency, resonance • Can create a control signal to adjust a lowpass filter cutoff frequency, and applying that lowpass filter to noise • Example: tutorial-a-05.csd sr = 44100 ksmps = 10 nchnls = 2 instr 102 iDur = p3 iTrem = p4 iFilterRate = p5 aEnvl aEnvl oscili .5, iTrem, 99 = aEnvl + .5 aNoise aNoise random -12000, 12000 = aNoise * aEnvl kCutoff kCutoff kCutoff oscili .5, iFilterRate, 99 = kCutoff + .5 = kCutoff * 8000 + 900 aPost lowpass2 outs endin aNoise, kCutoff, .85 aPost, aPost f 99 0 16384 10 1 i 102 0 2 6.2 .85 i 102 3 2 23 .65 i 102 6 2 45.6 .50 21.13. Csound: A Classic Synthesizer • A class subtractive synth sound with detuned oscillators, an LPF with modulated cutoff, and an ADSR envelope • Voltage controlled oscillator (vco): anlogue modelled digital oscillator aOsc • vco ADSR envelope kEnvel • amp, cps, waveShape, pulseWidth, functionTable adsr attack, decay, sustainLevel, release Example: tutorial-a-06.csd sr = 44100 ksmps = 10 nchnls = 2 instr 103 iDur = p3 iAmp = ampdbfs(p4) iPitch = cpsmidinn(p5) iFilterRate = p6 kCutoff kCutoff kCutoff oscili .5, iFilterRate, 99 = kCutoff + .5 = kCutoff * 4000 + 400 aOscA aOscB vco vco aPost kEnvl lowpass2 adsr outs endin f 99 0 16384 10 i 103 0 2 -12 i 103 3 2 -18 i 103 6 4 -24 iAmp, iPitch, 2, .5, 99 iAmp, iPitch*.499, 1, .5, 99 aOscA+aOscB, kCutoff, 1.2 .1*iDur, .2*iDur, .8, .2*iDur aPost*kEnvl, aPost*kEnvl 1 52 51 48 .5 1.2 3 21.14. Granular Synthesis: History • Isaac Beekman (1588-1637): 1616: corpuscular theory of sound: sound cuts air • 1947: Gabor proposes acoustical quanta: like photons for sound (1947) © Nature Publishing Group. All rights reserved. • 1960: Xenakis expands theory of screens and grains for creative sound production (Xenakis 1992) • 1978: Curtis Roads introduces software for Granular Synthesis (Roads 1978, 1996, p. 168, 2002) 21.15. Granular Synthesis: Concepts • Produce a stream of sounds with very short envelopes (10 to 200 ms) • Envelopes function like windows; multiple windows are often overlapped • Sounds may be derived from synthesized or sampled sources • Parameters are frequently randomly adjusted (spacing, amplitude, duration) © source unknown. All rights reserved. • Multiple streams are often combined combined • Extreme control and speed suggests a procedure idiomatic to computer-based synthesis 21.16. Granular Synthesis: Pitch Shifting and Time Stretching • The Eltro Information Rate Changer (1967) (c) Serendip LLC. All rights reserved. • Four playback heads 90 degrees apart on a cylinder; when spun can make continuous contact with the tape • By changing the direction and speed of the tape head rotation, could re-sample small bits of audio at a different speed without changing playback speed • By changing the tape speed and the tape head rotation speed, could alter tempo without altering pitch • Granular pitch/time shifting reads overlapping segments of an audio buffer, where each segment start position is consistent with the source playback speed, yet the reading of that segment can happen at a variable rate [demo/granularBasic.pd] 21.17. Grains in Csound • Numerous highly-specialized, advanced opcodes are available in Csound and other synthesis languages • “grain,” “granule,” (and more) for granular synthesis • An instrument that smoothly moves from min to maximum density, granulating an audio file loaded into a wave-table • Example: tutorial-a-07.csd sr = 44100 ksmps = 10 nchnls = 2 instr 104 iDur = p3 iDensityMin = p4 iDensityMax = p5 iSnd = 98 iBaseFq = 44100 / ftlen(iSnd) kDensity kGrainDur line line iDensityMin, iDur, iDensityMax .010, iDur, .030 kAmpDev aSrc line grain outs 0, iDur, 1000 16000, iBaseFq, kDensity, kAmpDev, 0, kGrainDur, iSnd, 99, .100 aSrc, aSrc endin f 99 0 16384 20 f 98 0 1048576 1 1 “sax.aif” 0 0 0 i 104 0.0 10 .05 20 i 104 11.0 10 35 200 21.18. Listening: Curtis Roads • Composer, computer musician, writer • Significant early work with granular techniques • Curtis Roads: “Now”: Line Point Cloud 21.19. Listening: Trevor Wishart • Sound mutations and transformations • Interest in vocal sounds and new notations (Wishart 1996) • Trevor Wishart, Red Bird, 1977 21.20. More Synthesis with Code • A variety of low-level frameworks for DSP in C and C++: STK, openAL • Numerous high-level languages related to Csound, often built in C/C++ • Text-based: SuperCollider, ChucK, Nyquist, Cmix, Cmusic • Graphic-based: PD/MaxMsp, Open Sound World, Reaktor 21.21. The Problem of Text • Systems like Csound are powerful, but may make exploration and experimentation difficult • Batch processing did not permit real-time, interactive systems • Signal graph (or signal network or patching) can be spread across multiple lines of text 21.22. Signal Processing Block Diagrams • Used in audio engineering • Used to plan voltage-controlled synthesis systems before execution • Used to illustrate unit generators and types of inputs and output in Music N languages • Examples: 21.23. MaxMSP/PD • Max is a visual programming paradigm • Many diverse implementations: MaxMSP, jMax, Pd • Emphasizes real-time control and signal flow design • Emphasizes processes more than data 21.24. MaxMSP/PD: History • 1979-1987: Miller Puckette studied with Barry Vercoe • 1982: Puckette releases Music 500 • 1985: Working on a dedicated digital audio processor, Puckette designs a new system, keeping the Music 500 control structure; names it Max after Max Mathews’s RTSKED (Puckette 1985) • 1987: Re-rewrites Max in C for Macintosh (Puckette 1988) • Max commercialized by David Zicarelli, fell through two companies, than reconsolidated at Cycling 74 • Puckette reprograms Max for IRCAM ISPW and NeXT Cube, and adds signal processing to Max (called Faster Than Sound (FTS)) • 1991: Max/FTS ported to other architectures • IRCAM version becomes jMax • Puckette reprograms system as Pure Data (PD), releases in 1997 as an open-source tool (Puckette 1997) • Zicarelli, after PD’s signal processing, creates Max Signal Processing (MSP) (Puckette 2002) • PD-Extended offers a complete package of PD tools for all platforms http://puredata.info/downloads 21.25. SuperCollider: History • Programming language and development environment for real-time signal processing • First released in 1996 by James McCartney (McCartney 1996; McCartney 1998) • 1999: version 2 released (Wells 1999) • In 2002 version 3 released as an open source project 21.26. SuperCollider: Concepts • Unit Generators are combined to produce SynthDefs • A server-based architecture: SynthDefs live on a server and send and receive messages and signals • A complete object-oriented language: create objects, manipulate, and reuse code • Designed for real-time performance and experimentation • Code can be executed piece by piece in the development environment • Under active development and supported by a robust community http://supercollider.sf.net 21.27. SuperCollider: Basic Patching • Can evaluate code interactively by selecting expressions and pressing Enter (not Return!) • Creating noise {WhiteNoise.ar(0.2)}.play • Enveloping noise with a sine envelope scaled {WhiteNoise.ar(0.2) * SinOsc.kr(4, mul:0.5, add:0.5)}.play • Oscillating rate of envelope applied to noise { var envRate; envRate = SinOsc.kr(0.3, mul:20, add:1.5); WhiteNoise.ar(0.2) * SinOsc.kr(envRate, mul:0.5, add:0.5); }.play • Applying a low-pass filter { var envRate, preFilter; envRate = SinOsc.kr(0.3, mul:20, add:1.5); preFilter = WhiteNoise.ar(0.2) * SinOsc.kr(envRate, mul:0.5, add:0.5); LPF.ar(preFilter, 900); }.play • Applying a low-pass filter with a cutoff frequency controlled by an oscillator; translating MIDI values to Hertz { var envRate, preFilter, cfControl; envRate = SinOsc.kr(0.3, mul:20, add:1.5); cfControl = SinOsc.kr(0.25, mul:0.5, add:0.5); cfControl = (cfControl * 70) + 50; preFilter = WhiteNoise.ar(0.2) * SinOsc.kr(envRate, mul:0.5, add:0.5); LPF.ar(preFilter, cfControl.midicps); }.play 21.28. SuperCollider: Creating SynthDefs and Sending Parameters • Most often, SynthDefs are created and sent signals or parameters from other processes • Create a SynthDef; create an envelope opened and closed by a gate; create LPF filtered noise; control the amplitude of the noise by the envelope; create a Task to loop through parameters for duration, sustain, and cutoff frequency scalar • Example: tutorial-b.rtf ( SynthDef(\noise, {|sus=2, ampMax=0.9, lpfCfScalar=20| var env, amp, gate, sigPrePan, cfControl; gate = Line.ar(1, 0, sus, doneAction: 2); env = Env.adsr(0.1*sus, 0.2*sus, 0.8, 0.1*sus, ampMax); amp = EnvGen.kr(env, gate); cfControl = SinOsc.kr(12, mul:0.5, add:0.5); cfControl = (cfControl * lpfCfScalar) + 40; sigPrePan = LPF.ar(WhiteNoise.ar(amp), cfControl.midicps); Out.ar(0, Pan2.ar(sigPrePan, 0.5)); }).send(s); r = Task({ var dur, sus, fq, delta; dur = Pseq([0.5, 0.5, 0.25], 6).asStream; sus = Pseq([0.2, 0.2, 0.2], 6).asStream; fq = Pseq([60, 30, 20, 40], 6).asStream; // midi pitch values while {delta = dur.next; delta.notNil } { Synth(\noise, [sus: sus.next, lpfCfScalar: fq.next]); delta.yield; } }); r.play() ) • Adding randomized panning control and cutoff frequency scalar • Example: tutorial-c.rtf ( SynthDef(\noise, {|sus=2, ampMax=0.9, lpfCfScalar=20, pan=0.5| var env, amp, gate, sigPrePan, cfControl; gate = Line.ar(1, 0, sus, doneAction: 2); env = Env.adsr(0.1*sus, 0.2*sus, 0.8, 0.1*sus, ampMax); amp = EnvGen.kr(env, gate); cfControl = SinOsc.kr(12, mul:0.5, add:0.5); cfControl = (cfControl * lpfCfScalar) + 40; sigPrePan = LPF.ar(WhiteNoise.ar(amp), cfControl.midicps); Out.ar(0, Pan2.ar(sigPrePan, pan)); }).send(s); r = Task({ var dur, sus, fq, delta, pan; dur = Pseq([0.5, 0.5, 0.25], 6).asStream; sus = Pseq([0.2, 0.2, 0.2], 6).asStream; fq = Pshuf([60, 30, 20, 40], 6).asStream; // midi pitch values pan = Pshuf([0, 0.2, 0.4, 0.6, 0.8, 1], 6).asStream; while {delta = dur.next; delta.notNil } { Synth(\noise, [sus: sus.next, lpfCfScalar: fq.next, pan: pan.next]); delta.yield; } }); r.play() ) 21.29. Live Coding • A performance practice of computer music that emphasizes the creation of code • Computer screens are projected while code is used to build-up musical parts • Software such as SuperCollider, Impromptu, and ChucK are used • Live Coding with aa-cell YouTube (http://www.youtube.com/watch?v=OBt4PLUv2q0)