rnnoise-nodejs

Node.js bindings to Xiph's RNNoise denoising C library.

RNNoise is a project showing how deep learning (Recurrent Neural Networks/RNNs) can be applied to noise suppression.

Announcement

Jan 2021:

I have been working on a new project that will bring this technology to you. A sneak peak at the new deep noise suppression and source separation model and audio output quality:

Watch on YouTube.

This is the 2020 state-of-the art noise suppression for real-time and offline use-cases. It works best in real-world environment where the background noise is low to medium. Of course it is not working well in super noisy environment like construction site. The model is suitable for many work-from-home environments such as home or cafe. We're currently working to bring this technology to desktop app for Mac OSX and Ubuntu (linux). Programming language SDK for Node.js and Go is also in the work. REST API is available (upon request) for integration with other use-cases.

Feb 2021:

The new project has 3 components:

1. PyTorch deep learning model
2. Node.js bindings for the PyTorch traced (JIT) model in C++
3. JavaScript port of PyTorch C++ library (libtorch)

Development of component 1 and 3 is completed.

Install

Node.js versions supported: 8, 10

Use

const rnnoise = require("rnnoise");

const denoisedBufLength = rnnoise.suppress(
  "babble_15dB.wav",
  "babble_15dB_dn.wav"
);

console.log(`Denoised buffer length: ${denoisedBufLength} bytes`);

API

Noise Suppression Functions

rnnoise.suppress(input: string, output: string)

suppress operates on 16-bit RAW audio format (machine endian) mono PCM files sampled at 48 kHz. The output is also a 16-bit RAW PCM file.

• input is a required string of the path to RAW PCM file input.
• output is a required string of the path to output RAW PCM file.

Developer

Working on project with submodules

We keep a rnnoise Git repo as a subdirectory in this Git repo. So, please clone this repo by using Git submodule:

git clone --recursive https://github.com/cedrickchee/rnnoise-nodejs.git

The code in this repository, including all code samples, is released under the MIT license.

Copyright (c) 2020 Cedric Chee