pipecat-firered-vad 0.1.0

FireRedVAD integration for Pipecat — SOTA streaming Voice Activity Detection supporting 100+ languages

pipecat-ai-fireredvad

A Pipecat integration for FireRedVAD — a SOTA industrial-grade streaming Voice Activity Detection model that supports 100+ languages and outperforms Silero-VAD, TEN-VAD, FunASR-VAD, and WebRTC-VAD on the FLEURS-VAD-102 benchmark.

Metric	FireRedVAD	Silero-VAD	TEN-VAD	WebRTC-VAD
F1 Score ↑	97.57	95.95	95.19	52.30
AUC-ROC ↑	99.60	97.99	97.81	—
False Alarm ↓	2.69	9.41	15.47	2.83

---

Requirements

• Python 3.10+
• pipecat-ai >= 0.0.90
• fireredvad (installed manually from GitHub — see setup below)
• Audio: 16 kHz, 16-bit mono PCM

---

Installation

1. Install this package

pip install pipecat-firered-vad

2. Install FireRedVAD

fireredvad is not on PyPI. Clone and install it from GitHub:

git clone https://github.com/FireRedTeam/FireRedVAD.git
cd FireRedVAD
pip install -r requirements.txt
export PYTHONPATH=$PWD:$PYTHONPATH

3. Download model weights

# via Hugging Face
pip install -U "huggingface_hub[cli]"
huggingface-cli download FireRedTeam/FireRedVAD \
    --local-dir ./pretrained_models/FireRedVAD

# or via ModelScope (recommended if you're in China)
pip install -U modelscope
modelscope download --model xukaituo/FireRedVAD \
    --local_dir ./pretrained_models/FireRedVAD

4. Configure environment

cp .env.example .env
# Edit .env and set FIREREDVAD_MODEL_DIR to your downloaded weights path

---

Quick Start

import asyncio
from dotenv import load_dotenv
import os

from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.runner import PipelineRunner
from pipecat.pipeline.task import PipelineParams, PipelineTask
from pipecat.audio.vad.vad_analyzer import VADParams
from pipecat_ai_fireredvad import FireVadAnalyzer

load_dotenv()

async def main():
    vad = FireVadAnalyzer(
        model_dir=os.environ["FIREREDVAD_MODEL_DIR"],
        params=VADParams(
            confidence=0.7,
            start_secs=0.2,
            stop_secs=0.5,
        ),
        speech_threshold=0.4,
        smooth_window_size=5,
    )

    # Pass the analyzer to your transport, e.g. DailyTransport:
    # transport = DailyTransport(..., vad_analyzer=vad)

asyncio.run(main())

---

Configuration Reference

FireVadAnalyzer constructor parameters

Parameter	Type	Default	Description
model_dir	str	—	Required. Path to the Stream-VAD model directory.
sample_rate	int	None	Must be 16000 if provided (enforced).
params	VADParams	None	Pipecat-level smoothing (confidence, start/stop secs).
use_gpu	bool	False	Run inference on GPU.
smooth_window_size	int	5	Frame-level confidence smoothing window inside FireRedVAD.
speech_threshold	float	0.4	Raw model threshold for speech vs silence.
pad_start_frame	int	5	Frames prepended at speech onset to avoid clipping.
min_speech_frame	int	8	Minimum consecutive frames before segment is confirmed.
max_speech_frame	int	2000	Maximum frames in a single speech segment.
min_silence_frame	int	20	Silence frames required before a segment ends.
max_buffer_frames	int	50	Ring buffer capacity (oldest frames evicted on overflow).

Audio requirements

FireRedVAD only accepts 16 kHz, 16-bit mono PCM. Convert other formats with:

ffmpeg -i input.wav -ar 16000 -ac 1 -acodec pcm_s16le output.wav

---

Environment Variables

Variable	Description
FIREREDVAD_MODEL_DIR	Path to the downloaded Stream-VAD model directory.
FIREREDVAD_USE_GPU	Set to 1 to enable GPU inference (default: 0).

See .env.example for a ready-to-copy template.

---

Related packages

• pipecat-firered-vad — this package
• pipecat-ai-ten-vad — TEN-VAD integration (same author)
• FireRedVAD on GitHub
• FireRedVAD on HuggingFace

---

Contributing

Pull requests are welcome. For major changes, please open an issue first.

1. Fork the repository
2. Create a feature branch: git checkout -b feature/my-change
3. Run linting: ruff check . && ruff format .
4. Run tests: pytest
5. Open a PR

---

License

Apache License 2.0 — see LICENSE for details.

FireRedVAD model weights are released under their own license. See the FireRedVAD repository for details.