PySTK: Python bindings for the Synthesis ToolKit

Python bindings for the Synthesis ToolKit in C++ (STK) by Perry R. Cook and Gary P. Scavone, powered by nanobind.

The bindings allow access to all classes that implement synthesis algorithms:

• stk::Effect and all derived classes
• stk::Filter and all derived classes
• stk::Function and all derived classes
• stk::Generator and all derived classes
• stk::Guitar
• stk::Instrmnt and all derived classes
• stk::Phonemes
• stk::Twang
• stk::Voicer

A lot of the functionality of other classes (File IO, Networking) can likely be substituted by common Python libraries, but might be added to the bindings as needed.

Installation

The bindings are available as prebuilt wheels for Linux, Windows and MacOS:

pip install synthesistoolkit

Alternatively, you can build the bindings from source:

git clone --recurse-submodules [email protected]:dsuedholt/pystk.git
cd pysdt && pip install .

Usage

Documentation for the Python bindings is not yet available, but the API is almost identical to the C++ API documented here.

The rawwaves folder is included in the distribution and its path is set in __init__.py.

tick(...) methods that deal with stk::StkFrames return a numpy.ndarray with the audio data.

import synthesistoolkit as stk
import numpy as np

# important: initialize sample rate before doing anything else
sr = 44100
stk.set_sample_rate(sr)

clnt = stk.Clarinet()
clnt.set_frequency(220)

# for classes that support control changes: see available control change numbers
print(stk.Clarinet.CONTROL_IDS)
# {'breath_pressure': 128,
#  'noise_gain': 4,
#  'reed_stiffness': 2,
#  'vibrato_frequency': 11,
#  'vibrato_gain': 1}

# control changes can be set directly
clnt.control_change(stk.Clarinet.CONTROL_IDS['reed_stiffness'], 1.8)

# or passed as a dictionary of audio rate parameters
controls = {
    stk.Clarinet.CONTROL_IDS['breath_pressure']: np.linspace(80, 40, num=sr),
    stk.Clarinet.CONTROL_IDS['vibrato_gain']: np.linspace(0, 90, num=sr),
}

audio = clnt.tick(sr, controls=controls)
print(audio.shape) # (1, 44100)

# can now use the audio in any way you like, for example:
import soundfile as sf
sf.write('clarinet.wav', audio[0, :], sr)