Our take
Maintaineda few commits in the last 90 days- Activity
- 30
- Community
- 50
- Issues
- 21
- Pull requests
- 43
+30 stars in the last 30 days
TEN VAD is a promising cross-platform option for latency-sensitive speech segmentation, with recent maintenance activity but no release in the past year and limited recent issue resolution.
Good fit if
- You need frame-level VAD for a conversational agent, ASR pipeline, or other real-time audio application.
- Your deployment can use its supported platforms and language interfaces, and can provide 16 kHz input.
- You want to evaluate the released ONNX path or use the prebuilt native libraries.
Look elsewhere if
- You need a recent versioned release as a prerequisite for adoption.
- Your audio pipeline cannot resample input to 16 kHz.
- Your organization cannot accept the license terms, which include additional conditions beyond the Apache 2.0 reference in the README.
All health signals
| Last commit | 2026-10-09 (23 days ago) |
|---|---|
| Commits, last 90 days | 3 |
| Releases, last 12 months | 0 (latest v1.0, 2025-07-11) |
| Contributors | 13 (top contributor: 57% of commits) |
| Issues closed, last 90 days | 0 |
| Pull requests merged, last 90 days | 1 |
| Project age | 18 months |
Checked on 2026-11-02 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
2,317Forks
182Change since Nov 16, 2026: +3 stars·0 forks
Overview
TEN VAD is a streaming voice activity detector for accurate, frame-level speech detection, designed to reduce latency in conversational voice agents and other real-time audio systems. It is built around the TEN VAD model, with an ONNX version and its preprocessing code released by TEN; the project also provides prebuilt native libraries and C interfaces.
The core library targets Linux, Windows, macOS, Android, and iOS, with interfaces for C, Python, Java, and Go, plus JavaScript support for WebAssembly. It expects 16 kHz audio, so other input rates need to be resampled. Teams can use the native prebuilt libraries or deploy the ONNX model with ONNX Runtime.
Key Features
- Produces frame-level speech activity results for streaming audio.
- Targets low-latency speech-to-silence detection for conversational agents.
- Provides a C library for Linux, Windows, macOS, Android, and iOS.
- Includes language interfaces or examples for Python, Java, Go, and JavaScript/WASM.
- Supports configurable frame hops, with 160- and 256-sample configurations described for 16 kHz audio.
- Offers an ONNX model and preprocessing code for deployment with ONNX Runtime.
- Includes an annotated test set and a script for reproducing precision-recall comparisons.
Use Cases
- Voice-agent developers who need to detect speech endings quickly to support turn-taking.
- Streaming ASR pipelines that need speech segments identified before or during transcription.
- Mobile and browser applications that need VAD without depending on a server-side audio pipeline.
- Developers evaluating VAD behavior on annotated audio or tuning the speech-probability threshold for a domain.
What you need
Detected in the repository
- Python (from setup.py)
- Automated checks on GitHub Actions
Getting Started
Clone the repository to access the libraries and examples:
git clone https://github.com/TEN-framework/ten-vad.git
See the README for platform-specific builds, language examples, and ONNX Runtime setup.
License in Plain Words
No standard license was detected. Without a license, the default copyright rules apply: you may read the code, but not reuse it. Check the repository or ask the authors.
Considerations
The project is marked maintained and has recent commits, but its latest release is more than a year old; the supplied health data also shows no issues closed in the recent 90-day period. The README documents prebuilt-library and ONNX deployment paths, but Python requires NumPy and the ONNX path requires ONNX Runtime. The repository's license description notes additional conditions and separate notices for derived code, so review the included LICENSE and NOTICES files before distribution.
Found this useful?
Share it with someone who would like ten-vad.