yaapt.torch

Implementation of the YAAPT pitch tracker in PyTorch.
This version relies solely on PyTorch, does not hog all CPU threads (making it possible to use in a dataloader), is torch.jit-able, and is batch compatible.

Setup

For using (no development required)

pip install yaapt.torch

To install for development, clone the repository, and then run the following from within the root directory.

pip install -e .

Usage

import torchaudio, yaapt_torch
audio, sr = torchaudio.load("audio.wav")

yaapt_opts = {
    "sr": sr,
    "frame_length": 35.0,
    "frame_space": 20.0,
    "nccf_thresh1": 0.25,
    "tda_frame_length": 25.0,
}

pitch = yaapt_torch.yaapt(
    audio, # (Batch, feature) torch.tensor input
    yaapt_opts,
)
print(pitch) # (Batch, F0) torch.tensor output

Credits

• The original NumPy/SciPy/Python program was written by Bernardo Jose B. (https://github.com/bjbschmitt).
• The original MATLAB program was written by Hongbing Hu and Stephen A. Zahorian (http://www.ws.binghamton.edu/zahorian/yaapt.htm).

This repository aims to wrap up these implementations in easy-installable PyPi packages, which can be used directly in PyTorch based neural network training.