PyWORLD Documentation

Keep these constraints in mind when using PyWORLD:

• Sampling Rate: WORLD is designed for speech sampled at ≥ 16 kHz. Applying it to 8 kHz speech will fail.
• Low SNR: If the Signal-to-Noise Ratio (SNR) is low, use the harvest algorithm instead of dio for better pitch extraction.

The easiest way to install the PyWORLD wrapper is using pip.

pip install pyworld

To build from source, clone the repository, initialize submodules (which includes the C++ WORLD Vocoder), and install the requirements. It is recommended to use virtualenv or conda for this process.

git clone https://github.com/JeremyCCHsu/Python-Wrapper-for-World-Vocoder.git
cd Python-Wrapper-for-World-Vocoder
git submodule update --init
pip install -U pip
pip install -r requirements.txt
pip install .

After installation, you can verify that everything is working correctly by running the provided demo script. This will generate results in the test/ directory.

cd demo
python demo.py

Common issues and solutions:

• Cython version: If building fails, upgrade Cython to version 0.24 or higher. If pip install Cython does not work, download it from cython.org and run python setup.py install.
• Matplotlib in remote environments: If running demo.py on a remote Linux server, you may need to use the 'Agg' backend:

  import matplotlib
  matplotlib.use('Agg')

• Missing sndfile: If you see library not found: sndfile, install it via apt-get install libsoundfile1.
• Windows Installation: If you encounter issues on Windows, consider using WSL (Windows Subsystem for Linux).
• Mac Users: You might need to set the deployment target during installation: MACOSX_DEPLOYMENT_TARGET=10.9 pip install ..

PyWORLD provides low-level functions to parameterize speech into three components: f0 (pitch contour), sp (harmonic spectral envelope), and ap (aperiodic spectral envelope). You can then use these to synthesize speech.

import pyworld as pw

# 1. Extract pitch
_f0, t = pw.dio(x, fs)    # raw pitch extractor
f0 = pw.stonemask(x, _f0, t, fs)  # pitch refinement

# 2. Extract spectral envelopes
sp = pw.cheaptrick(x, f0, t, fs)  # extract smoothed spectrogram
ap = pw.d4c(x, f0, t, fs)         # extract aperiodicity

# 3. Synthesize speech
y = pw.synthesize(f0, sp, ap, fs)

The wav2world utility function provides a high-level way to convert speech into f0, sp, and ap features using default arguments.

import pyworld as pw

# Convert speech into features (using default arguments)
f0, sp, ap = pw.wav2world(x, fs)