audio
Data manipulation and transformation for audio signal processing, powered by PyTorch
What is audio?
Data manipulation and transformation for audio signal processing, powered by PyTorch
Overview of audio
What it is
torchaudio is an audio library for PyTorch that applies PyTorch to the audio domain. It provides GPU acceleration, trainable features through the autograd system, and consistent tensor naming. It is primarily a machine learning library, not a general signal processing library. All computations are performed through PyTorch operations, making it a natural extension of PyTorch. The library is currently in a maintenance phase, with some user-facing features removed starting from version 2.9 to reduce redundancies with the PyTorch ecosystem and focus on processing audio data for machine learning.
Key features
- Dataloaders for common audio datasets
- Audio and speech processing functions, including forced alignment
- Common audio transforms: Spectrogram, AmplitudeToDB, MelScale, MelSpectrogram, MFCC, MuLawEncoding, MuLawDecoding, Resample
- Compliance interfaces that run code using PyTorch aligned with other libraries, such as Kaldi (spectrogram, fbank, mfcc)
Use cases
torchaudio is designed for machine learning tasks involving audio data, such as speech recognition, self-supervised learning, and audio processing. It can be used to load and preprocess audio datasets, extract features like spectrograms and MFCCs, and apply transforms for model training. The library's integration with PyTorch allows for GPU-accelerated computation and gradient-based optimization of audio features.
Getting started
bashpip install torchaudioTorchAudio 2.11 requires torch 2.11 or newer, including all future releases. It does not pin torch to a specific version. For full installation and build instructions, and for the compatibility matrix of older releases, please refer to the official installation page.
License and pricing
The library is open source under the BSD-2-Clause license. Pricing is not stated; it is an open-source project.
Based on the project's public README and repository metadata.