Voice Activity Detection for the Browser

This package aims to provide an accurate, user-friendly voice activity detector that runs in the browser. Currently, it runs Silero VAD [1] in the browser using ONNX Runtime Web.

A demo is hosted at vad-demo-script.vercel.app. The source code for the demo can be found here. A separate demo showing how to use the VAD with a bundler like webpack can be found here.

The API works as follows:

1. Create the VAD object with a line such as

   const myvad = await vad.MicVAD.new(options)

   options can include any of the parameters defined here. It essentially consists of callbacks that run on every audio frame, whenever a speech start is detected, whenever speech ends, etc, as well as parameters that control the voice activity detection algorithm.

2. Start and pause the VAD object as needed with myvad.start() and myvad.pause(). The object starts in the paused state.

References

[1] Silero Team. (2021). Silero VAD: pre-trained enterprise-grade Voice Activity Detector (VAD), Number Detector and Language Classifier. GitHub, GitHub repository, https://github.com/snakers4/silero-vad, hello@silero.ai.