facebookresearch/encodec
A neural audio codec using deep learning for high-fidelity audio compression at multiple bandwidths.
Not currently ranked — collecting fresh signals.
star history
State-of-the-art neural audio codec presented in the High Fidelity Neural Audio Compression paper. Uses convolutional neural networks, LSTM encoders, and residual vector quantization to compress audio at various bitrates (1.5 to 24 kbps). Supports both causal 24 kHz monophonic audio and non-causal 48 kHz stereophonic music. Includes an entropy model based on a small transformer that can further compress representations by up to 40% without quality loss, along with a novel multiscale complex spectrogram discriminator for training.
Frequently asked
- What is facebookresearch/encodec?
- A neural audio codec using deep learning for high-fidelity audio compression at multiple bandwidths.
- Is encodec open source?
- Yes — facebookresearch/encodec is open source, released under the MIT license.
- What language is encodec written in?
- facebookresearch/encodec is primarily written in Python.
- How popular is encodec?
- facebookresearch/encodec has 4k stars on GitHub.
- Where can I find encodec?
- facebookresearch/encodec is on GitHub at https://github.com/facebookresearch/encodec.