Skip to content
tiny spectral synthesizer with livecoding support
C++ Shell
Branch: master
Clone or download

Latest commit

Fetching latest commit…
Cannot retrieve the latest commit at this time.


Type Name Latest commit message Commit time
Failed to load latest commit information.

A tiny C++ live-coded overlap-add (re)synthesizer for Linux. Things that you can do easily with tinyspec include:

  • create novel audio effects using FFT, phase vocoders and more, and control them with OSC
  • create synthesizers
  • granular synthesis
  • bytebeats (time and frequency domain)
  • control other software with OSC
  • use these synthesizers and effects with DAWs, other synthesizers, etc using JACK
  • do all of this in a live performance (with some caveats)

What is an overlap-add (re)synthesizer? You define a function which is called periodically to process a frame of audio. The rate at which the function is called, as well as the length of a frame can be dynamically adjusted. If multiple frames overlap in time, some of the input that they receive will be the same, and their outputs will be mixed.

For example, you could set the frame size to 1 and the hop size to 1 if you want to process a single sample at a time. You can do granular synthesis of a wave at a specific frequency by e.g. setting the frame size to 100 and the hop to 440Hz. Or, you could set the frame size to 1024 and the hop size to 256 (ratio 1/4) for standard use of a phase vocoder. It's even possible to dynamically adjust the frame size and the hop according to some function. The amount of latency is automatically adjusted by inserting silence to the input as needed.

Visually, in the following diagram the hop is 3 and the frame size starts at 7 but increments by 1 each frame. The audio output from the frames which overlap vertically will be added together before sending it out via JACK.

F |-----|
r    |------|
a       |-------|
m          |--------|
e             |---------|
s                |----------|
|                   |-----------|
|                      |------------|
V                         ...

You probably wouldn't want to literally do this, because the frames would quickly grow to an unwieldy length and the output would likely clip from having too many frames overlap at once. So, you do have to be a bit careful about what you're doing. Also, be aware that things like division by zero and null pointer dereferences will crash your program. Hence why you should be particularly vigilant if you decide live code this in a performance.

This is a fork of tinyspec, which was an experiment in trying to make the smallest useful FFT synthesizer. This version adds live-coding support via cling, among many other handy features. Cling is quite large and only officially supports Ubuntu (but probably works elsewhere), so if you don't want to use it, you can still use this to compile your code into a standalone executable, without live-coding support.


Download an appropriate cling package from and extract it to a directory called cling_bin. (If you're using Arch, the Ubuntu 18 package gives some warnings but basically seems to work.) This step is not necessary if you only want to use standalone mode (see below).

JACK Audio Connection Kit is used for audio output, and FFTW3 is required for synthesis.

Standalone mode

The easiest way to get started is using standalone mode. For example, try

$ ./standalone hacks/readme.cpp

Then in a separate terminal, connect the program to your speakers by running

$ jack_connect tinyspec_cmd:out0 system:playback_1
$ jack_connect tinyspec_cmd:out1 system:playback_2

See hacks/connect.cpp for examples of how you can make your program automatically connect inputs/outputs. You can also do this through a JACK GUI such as Catia.

Live-coding mode

Running the application in live coding mode starts a server which listens for code written to a named pipe. You can start as many servers as you want and route their inputs/outputs using JACK. To do this, you can use a JACK GUI such as Catia. Alternatively see hacks/connect.cpp for examples of how you can make your program automatically connect inputs/outputs.

Compile tinyspec by running ./compile.

To run the server do:

$ ./tinyspec /tmp/ts-example

Where /tmp/ts-example is the pipe to create for executing code from. If you run multiple instances they must have unique pipe files. If all goes well you should see


There may be some warnings from cling afterwards, which may or may not be safe to ignore! (cling is very buggy) You won't hear anything yet for two reasons: the application is not connected to your speakers (do this through JACK) and we haven't sent it any code yet.

Editor setup

The send program can be invoked on the command line to execute code on the server. It takes the path to the server's command pipe as an argument.

For example try

$ ./send /tmp/ts-example
cout << "Hello World!" << endl;

It's probably easy to set up your editor to invoke send with the current selection or block. For example, the following vim commands map F2 to execute the current paragraph of code.

:map <F2> mcVip::w<Home>silent <End> !./send /tmp/ts-example<CR>`c
:imap <F2> <Esc>mcVip::w<Home>silent <End> !./send /tmp/ts-example<CR>`ca

Check out doc/vscode.txt for a basic method of using tinyspec with vscode.

If you configure another editor, please contribute it here so other users can benefit!


For example, try executing the following code. You can also find this file as hacks/readme.cpp

// Called periodically to fill up a new buffer.
// in and out are audio sample buffers
// n is the number of samples in the frame
// t is the time in seconds since the beginning of playback.
set_process_fn([&](WaveBuf& in, WaveBuf& out, int n, double t){
    FFTBuf fft;
    fft.resize(out.num_channels, n);
    // Loop over frequency bins. Starting at 1 skips the DC offset.
    for (int c = 0; c < 2; c++) {
        for (int i = 1; i < n; i++) {
            cplx x = sin(i*pow(1.0/(i+1), 1.0+sin(t*i*M_PI/8+c)*0.5))*25 // Some random formula
                /pow(i,0.9); // Scale magnitude to prevent loud high frequency noises.
            fft[c][i] = x; // Fill output buffer
    frft(fft, fft, -1.0); // Perform in-place inverse FFT
    out.fill_from(fft); // Copy real part to output
    window_hann(out); // Apply Hann window
    next_hop_ratio(4096); // Set FFT size for the next frame

Here, cplx is just an alias for std::complex<double>.

See synth.h for info about working with buffers and the built-in audio processing library.

Note the call to next_hop_ratio at the end---this is one of 3 built in functions that can be used to manipulate the parameters of the FFT synthesizer.

void next_hop_ratio(uint32_t n, double hop_ratio=0.25): Set the FFT size for the next frame to n and advance the output by n*hop_ratio samples.

void next_hop_samples(uint32_t n, uint32_t hop): Set the FFT size for the next frame to n and advance the output by hop samples.

void next_hop_hz(uint32_t n, double hz): Set the FFT size for the next frame to n and advance the output such that the rate of frame generation is hz Hz. That is, advance by RATE/hz samples.

You can also call void set_num_channels(size_t in, size_t out); to set the number of input and output channels to use.

Finally, if you notice a lot of latency, you can force the synthesizer to discard buffered data and skip to the present moment by calling void skip_to_now();

See the hacks directory for some other examples. Even more examples---which require adaptation for this fork---are available as part of tinyspec.

OSC support

There is currently experimental support for sending/receiving Open Sound Control messages. See hacks/osc.cpp and hacks/tidal.cpp for details.

You can’t perform that action at this time.