# digital music making tools
* analog hardware synthesizers (keyboards, modular racks, sequencers)
* daw software and sequencers for arranging instruments, recordings, and effects (e.g. ardour, cubase)
* modular environments and patchers (e.g. reaktor, max, puredata)
* plugin formats for instruments and effects (lv2, vst)
* midi and control-surface integration for hardware–software interplay
* scripting or programming languages to generate/process sample arrays and export audio files

# various notes
* harmonic: a waveform with frequency equal to a positive integer multiple of its fundamental ([wikipedia](https://en.wikipedia.org/wiki/Harmonic))
* time series: data points indexed or graphed in time order, often at equal intervals ([wikipedia](https://en.wikipedia.org/wiki/Time_series))
* clipping: limiting a signal’s amplitude to a defined maximum
* sound program as potential: code defines a space of sound events, with literal arguments acting as seeds for generation
* harmonicity–noise continuum: sounds range from pure harmonic spectra to full noise
* pattern complexity: emerges from layered, time-based mutations and transformations rather than mere variety of subsequences

# autocorrelation vs absolute differences
* autocorrelation gives a degree of difference with large local difference being pronounced to differentiate it from many small differences.
* correlation is -1 to 1
* correlation only tests linear dependence. (1 2 3) and (4 5 6) are 1
* mean absolute difference goes to infinity

# example framework choices for music creation
## principles
* avoid post-processing
  * modification of float values can be lossy
  * control as much as possible by initial arguments
  * prefer additive to subtractive processes
* times and offsets are specified in number of samples in cases where hertz and seconds do not map exactly to sample counts
* use wavelengths whose extremes fall exactly on samples
* use a high resolution for processing of modulation. ideally sample resolution
* processing is done for multiple channels with one array per channel
* sample format is variable (32 or 64 bit float, ideally more)
* sample rate is variable (low for fast processing). target sample rate is 96000. a sine of 12000 hz could then still be approximated with 8 samples
* tend to give up realtime processing for higher precision and more content. render to file, which is common in 3d modeling software
* when using random number generators, specify random seeds to keep the output repeatable

## toolset
* filtered noise generator whose amplitudes and filter values can be controlled by paths. customisable noise generator
* sine generator that keeps a continuous phase when frequency modulated and whose amplitudes and wavelength can be controlled by paths. sine(amp, frq, phs)
* file output with creation of wave files (ideally au, an extremely simple and capable file format)
* random number generator with customisable probabilities
* paths
  * interpolation between points
  * linear and rounded (bezier) interpolation methods
  * segment types are linear, bezier, sub-path or constant
* sequencer
  * for partials and whole songs
  * calls functions that write to sample buffers at specified times
  * called with a time offset, output buffers and an event list
  * sequencer calls can be nested to build composite sounds, patterns and riffs
  * processing can be done in parallel (top-level use) or serial (nested use)
  * event
    * have the properties (state start end function)
    * multiple events can be grouped into one event
    * output can be post-processed
    * output can be cached
* filters
  * windowed-sinc band-pass that optimises to low-pass and high-pass at the edges
  * state-variable filter with low/high/band pass output for cheaper filter operation
  * all filters must be able to filter a series of sample arrays seamlessly