Repolode
Submit
FireRedTeam

FireRedTeam/FireRedVAD

Open-source voice activity detection (VAD) and audio event detection (AED) for speech, singing and music across 100+ languages, with streaming and non-streaming modes.

View on GitHub
Stars
549
Forks
38
Open issues
16
Last commit
6 months ago

What is it?

FireRedVAD detects where speech occurs in audio and, separately, labels audio events such as speech, singing and music. It supports non-streaming VAD, streaming VAD for real-time input, and non-streaming AED, with detection covering more than 100 languages.

The project provides a pip package with a `fireredvad` command-line tool and a Python API (`FireRedVad`, `FireRedStreamVad`, `FireRedAed`). Input must be 16kHz 16-bit mono PCM WAV, and ffmpeg is suggested for conversion. Models are downloadable from Hugging Face or ModelScope, and an NCNN runtime is available for multiplatform deployment.

The models are built on DFSMN architectures. On the FLEURS-VAD-102 benchmark that the project created, non-streaming VAD reports a 97.57% F1 score, compared with figures listed for Silero-VAD, TEN-VAD, FunASR-VAD and WebRTC-VAD.

Key features

  • 01Non-streaming and streaming voice activity detection.
  • 02Non-streaming audio event detection for speech, singing and music.
  • 03Command-line tool and Python API for batch and programmatic use.
  • 04Supports 100+ languages according to its evaluation set.
  • 05NCNN runtime available for multiplatform deployment.

Use cases

  • Trimming silence and segmenting speech before passing audio to an ASR system.
  • Real-time voice detection in streaming audio pipelines.
  • Labeling speech, singing and music regions in media files.
  • Preprocessing multilingual recordings for speech datasets.

Topics

#voice-activity-detection#audio-event-detection#speech#audio#python#streaming#asr-preprocessing#aed#audio-event-classification#sound-event-detection

FAQ

What is FireRedVAD?
Open-source voice activity detection (VAD) and audio event detection (AED) for speech, singing and music across 100+ languages, with streaming and non-streaming modes. FireRedVAD detects where speech occurs in audio and, separately, labels audio events such as speech, singing and music. It supports non-streaming VAD, streaming VAD for real-time input, and non-streaming AED, with detection covering more than 100 languages.
Is FireRedVAD free to use? What license does it have?
FireRedVAD is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is FireRedVAD still maintained?
Updated within 6 months. The last commit to FireRedVAD was on May 6, 2026.
What can I use FireRedVAD for?
Trimming silence and segmenting speech before passing audio to an ASR system. Real-time voice detection in streaming audio pipelines. Labeling speech, singing and music regions in media files. Preprocessing multilingual recordings for speech datasets.
Where can I get FireRedVAD?
The source code and installation instructions are on GitHub: https://github.com/FireRedTeam/FireRedVAD

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects