pYIN algorithm written in Rust

This crate provides a pitch estimate for each frame of the audio signal and a probability that the frame is a voiced region.

The implementation is based on librosa. For easy translation from Python + Numpy to Rust, the implementation is written on top of ndarray crate.

Build & Run

You can use this both as an executable binary and as a library (C shared library and Rust library).

As an executable binary

cargo run --release <input_file> <output_npy_file> <fmin> <fmax> --frame_ms <frame length in miliseconds>

or

cargo build --release
./target/release/pyin <input_file> <output_npy_file> <fmin> <fmax> --frame_ms <frame length in miliseconds>

Note

• Supported audio files: the same as Creak crate.
  • Multi-channel audio files are supported.
• output file: npy file contains the output ndarray with
  • shape: (4, no. of channels in input audio, no. of frames)
  • [0, :, :]: timestamp [sec]
  • [1, :, :]: f0 array [Hz]
  • [2, :, :]: voiced flag(1.0 for voiced, 0.0 for unvoiced) array
  • [3, :, :]: voiced probability array
• If "-" is used as the output filename, the app will send output data to stdout.

Example using pYIN as a C shared library

The example is in test/test.c. To build and run it with GCC,

./compile_test.sh
LD_LIBRARY_PATH=target/release ./test_pyin

Using pYIN as a Rust library

Add the following to your Cargo.toml:

[dependencies]
pyin = "1.0"

TODO

• Input from stdio
• More options supported by command-line arguments