FireRedTeam/FireRedChat-pvad
By FireRedTeam · ↓ 0 · ♥ 12 · Updated 2025-09-22
What's in this repo
This page lists the files in FireRedTeam/FireRedChat-pvad, including pvad.onnx.
- 1 ONNX models (.onnx files)
- 12 files, including 4 .ckpt, 2 .md, 1 .onnx, 1 .json, 1 .yaml, 1 .txt
- Repository size ~88.8 MB
- Last updated 2025-09-22
ONNX files
| File | Size | Download |
|---|---|---|
| pvad.onnx | 3.8 MB | Download |
How to use
These ONNX files can be loaded with ONNX Runtime. To inspect a model graph, upload the file to the Netron viewer. Download links point to the original files on Hugging Face.
README
FireRedChat-pVAD
Demo • Paper • Huggingface
Descriptions
FireRedChat's personalized Voice Activity Detection (pVAD) model, an open-weight model for detecting voice activity with speaker embedding updates.. LiveKit plugin available here
- Supports speaker embedding updates for improved voice activity detection.
- The plugin requires a compatible LiveKit Agents fork or modification to include
update_speakercall for the first user utterance.
Roadmap
- [x] 2025/09
- [x] Release the pVAD model weights and LiveKit plugin.
Usage
For inference, please use the LiveKit plugin. Install and configure as follows:
from livekit.plugins import fireredchat_pvad as pvad
def prewarm(proc: JobProcess):
proc.userdata["vad"] = pvad.VAD.load(activation_threshold=0.5)
# After the first utterance (or when primary speaker switches based on RMS), call VADStream's update_speaker() to update speaker embedding.
License
The model weights and plugin code are licensed under the Apache-2.0 license.
Acknowledgment
- Speaker embedding model: speechbrain/spkrec-ecapa-voxceleb
Model info
Created: 2025-09-15
Updated: 2025-09-22
View on Hugging Face