pywavers 0.1.0

A mixed Rust and Python crate/package for reading and writing WAVE files. Exposes the Rust crate Wavers to Python.

PyWaveRS - WaveRs but for Python

WaveRs (pronounced wavers) is a Wav file reader/writer written in Rust and designed to fast and easy to use. WaveRs is also available in Python through the PyWaveRs package.

Getting Started · Benchmarks

---

Getting Started

Benchmarks

dtype	PyWavers	SoundFile	SciPy	TorchAudio (Sox)	TorchAudio (SoundFile)	TensorFlow Audio-IO
int16	18.0455	31.505	16.8316	195.954	191.826	49.0038
int32	35.8505	141.546	nan	nan	nan	nan
float32	43.3757	203.17	nan	228.13	225.659	nan
float64	81.4131	206.821	nan	324.17	315.95	nan

All times are in microseconds μs and are the time taken (on average) to load a single file from the set of test files. Timings were only taken when the library could load and automatically convert the file to another type of sampled audio data. Only PyWavers and Soundfile have this functionality built in. For the cases of f32 and f64 the conversion involves just casting to the other type therefore the times are included where applicable.

The NaNs present in the benchmarks table incidate the the library could not automatically load and convert to that data type. For example, SciPy loads the native encoding of the file, it isn't that it can't load a file, it just can't convert it to another type of sampled audio data. The same occurs with Tensorflow. PyTorch has the options of normalizing or not normalizing the data (as f32 or native encoding), but it doesn't have the option to convert to another type of sampled audio data, such as between i16 and i32.

The wav files included in the test set are between 7 and 15 seconds long. They are all encoded as PCM-16. The benchmarks were taken using Wavers v1.4.3.

Benchmark results for reading PCM-16 encoded wav files.