Softsynth engine for tiny & not-so-tiny devices…
- Demo set with 40+ tracks on Soundcloud
- STM32F746 MIDI synth (live performance, 2016-01-31, Youtube)
- STM32F746 synth (live recording, 2016-01-28, Youtube)
Please visit project log @ Hackaday for interim updates, including a PDF with more project details. Thanks!
This repo currently contains two different experimental & easily extensible implementations of a modular soft synth engine, both written in portable C11 and originally developed for the STM32F4/F7 platform. However, for practical reasons (repo size, easier debugging and dev workflow), both versions currently included here were prepared without STM specifics, have been tested on OSX & Linux and include several examples. Demo file size after stripping is between ~22-40KB (for 32bit, bit larger for 64bit).
The sub-directories contain:
src
- an earlier (2nd or 3rd) iteration of the CTSS synth engine, using block based operators & node DAG (directed acyclic graph):- 10 oscillator types (incl. formant & karplus-strong, polyBlep antialiasing)
- flexibile LFO setup (global LFO & arbitrary per voice)
- filters (biquad, IIR LP/HP/BP, formant)
- ADSR
- foldback distortion
- buffer ops (sum, product, product sum)
- mono / stereo delay lines w/ feedback
- mono / stereo mixdowns (int16 & float)
examples
- examples for this version of the synth (using Portaudio, see below)
(not embracing ANS Forth standard here):
The supplied Forth-like VM & core vocab is general purpose, is configurable (e.g. support for floating point & automatic inlining) and can also be used independently, outside this project. In fact, the synth DSL related aspects are implemented as add-ons and can easily be removed.
src_vm
- actual stack VM, interpreter & compiler w/ basic inlining, REPL, core Forth dictionary & basic synth operators (still need to port more fromsrc
)examples_vm
- Forth examples (see below)
Requirements:
The library itself has no 3rd party dependencies. Some of the examples however use these additional libraries:
- ct-head (included as submodule)
- Portaudio (on OSX can be installed via
brew
) - libncurses (part of OSX & Linux)
git clone --recursive https://github.com/thi-ng/synstack
cd synstack
premake5 gmake
make -f ctss_old.make config=release_sse
make -f ctss_vm.make config=release_sse
Use the above build instructions to build all components & examples.
Almost all demos in examples
implement a simple randomized sequence
playback (some with multiple voices) and are using the same
interactive controls:
Key | Description |
---|---|
- | dec delay feedback |
+ | inc delay feedback |
[ | pitch -2 semitones |
] | pitch +2 semitones |
; | pitch -12 |
‘ | pitch +12 |
q | quit |
The two exceptions are:
- synth_keys - play a new note with each keypress (fixed sequence,
q
to quit) - synth_render - non-interactive demo, creates /sound.wav file (runtime ~20secs)
# if you followed above build instructions, the demos can be launched like:
bin/release/synth_spiral
Use the command below to launch the VM with input file. Also see
example source code. In this mode the VM goes straight into playback
mode and is non-interactive (hit Return
to quit).
bin/release/repl examples_vm/bass-seq.fs
To launch the Forth REPL in interactive mode:
rlwrap bin/release/repl
There’ll be alot of output initially (inlining statistics by the Forth compiler), but no visible input prompt. Try some of the following (hit return after each line, comment lines are prefixed with ‘\’)…
\ evaluate expression
\ forth uses RPN, the '.' word pops & prints the top stack item
\ forth is a hybrid between interpreted & compiled execution
1 2 + .
\ 3
\ define new word (here to calculate a * b + c)
\ the (...) form is a stack effect comment:
\ here it means the word consumes 3 items from the stack
\ and after execution a new item (result) is on the stack
: madd ( a b c -- x ) -rot * + ;
3 5 10 madd .
\ 25
\ the VM differentiates between int & float based math
\ float ops are prefixed with 'f'
: maddf -rot f* f+ ;
3.f 5.f 10.f maddf .f
\ 25.0
\ store values in "variables" (labelled addresses)
23 var> foo !
\ get var value
foo @ .
\ 23
\ by default the VM includes the dictionary of synth related words
\ (defined in /src_vm/synth_dict & various C files in /src_vm)
\ for example note names map to their frequency
A4 .f
\ 440.000000
\ convert frequency to ratio of sample rate
A4 hz .f
\ 0.062689
\ allocate memory for note/chord lookup table
C3 E3 G3 A3 4 dup allot dup var> chord ! swap ds>dict!
\ define word to randomly choose note from chord
: rand-chord-note ( -- freq ) 0 4 rand chord @ + @ ;
\ call several times
rand-chord-note .f
rand-chord-note .f
rand-chord-note .f
\ disassemble word
see> rand-chord-note
\ 0x000004da :0x0301a88a docolon
\ 0x000004db :0x00000004 lit 0x00000000
\ 0x000004dd :0x00000004 lit 0x00000004
\ 0x000004df :0x00000081 rand
\ 0x000004e0 :0x00000004 lit 0x000004d3
\ 0x000004e2 :0x0000002a @
\ 0x000004e3 :0x0000004b + 0x0000002a (0x0000050e )
\ 0x000004e5 :0x00000002 ret
\ display data stack contents with .s
1 2 3 .s
\ display VM stats
.vm
\ display list & addresses of all words currently defined
\ also see /src_vm/core_dict.h how some control structures
\ are defined in forth itself
.words
\ display hexdump of VM memory
.mem
Name | Role | Website |
Karsten Schmidt | initiator & principal developer | thi.ng |
This project is open source and licensed under the Apache Software License 2.0.