RNNoise
What it is
RNNoise is a free, open-source noise suppression library based on a recurrent neural network. It is intended for use as a library and includes a command-line demo. The demo processes mono, 16-bit raw PCM audio at 48 kHz; its input and output are raw files rather than WAV. The README gives an autotools build sequence using autogen.sh, configure, and make. For CPU optimization, the project recommends AVX2 and describes AVX2 or SSE4.1 runtime CPU detection as build options. RNNoise can use its included default model or load a custom model through its API from a file, filename, or memory buffer. The README also describes a smaller alternative model with increased sparsity. Training requires clean speech and noise data sampled at 48 kHz in 16-bit PCM format. The repository identifies its license as BSD-3-Clause and lists Linux as its platform. It describes GitHub as a convenience copy and points to GitLab for the latest source. Xiph directs people to mailing lists or IRC channels for contact.
For and against
For
- Open-source under a BSD-3-Clause license
- Includes a command-line demo
- Can use a default or custom model
- Build options include AVX2 or SSE4.1 runtime detection
Against
- Listed platform is Linux
- Demo accepts mono 16-bit raw PCM at 48 kHz
- Demo input and output are raw files, not WAV
- Training requires clean speech and noise data in a specified format
The verdict
RNNoise is a focused library with a demo, model options, and documented build and CPU-optimization choices. Its Linux listing and raw PCM demo requirements are important constraints to check first.
Who it suits
RNNoise may suit developers looking for a Linux noise suppression library who can work with raw PCM input and build from source. Its API also allows loading a custom model.
Plans and prices
RNNoise
BSD-3-Clause license · no paid plans stated
Read 4 Oct 2026 · github.com
On the record
- Build
- The README gives an autotools build sequence using autogen.sh, configure, and make.github.com · read 4 Oct 2026
- CPU optimization
- The project recommends AVX2 and describes AVX2 or SSE4.1 runtime CPU detection as build options.github.com · read 4 Oct 2026
- Distribution
- The README says the GitHub repository is a convenience copy and identifies GitLab as the location of the latest source.github.com · read 4 Oct 2026
- Input format
- The demo processes mono 16-bit raw PCM at 48 kHz and its input and output are raw files, not WAV.github.com · read 4 Oct 2026
- License
- The repository identifies its license as BSD-3-Clause.github.com · read 4 Oct 2026
- Models
- RNNoise can use its included default model or load a custom model from a file, filename, or memory buffer through its API.github.com · read 4 Oct 2026
- Purpose
- RNNoise is a noise suppression library based on a recurrent neural network.github.com · read 4 Oct 2026
- Security
- The opened project pages state a software license but do not describe security certifications or compliance.github.com · read 4 Oct 2026
- Smaller model
- The README describes an alternative little model with increased sparsity.github.com · read 4 Oct 2026
- Support
- Xiph directs people to its mailing lists or IRC channels for contacting the foundation and community.xiph.org · read 4 Oct 2026
- Training
- Model training requires clean speech and noise data sampled at 48 kHz in 16-bit PCM format.github.com · read 4 Oct 2026
- Use
- The project is intended to be used as a library and includes a command-line demo.github.com · read 4 Oct 2026
