Voice Activity Detection — ai-coustics
Real-time, noise-robust turn-taking for voice agents.
Production VAD models by ai-coustics that decide who is speaking when directly on noisy microphone input — designed to work without a separate denoiser, in parallel with or independent of speech enhancement. Outperforms Silero VAD in accuracy, balance, and reliability; 30 ms latency, no GPU needed, no ONNX dependency. Real-time audio intelligence that makes Voice AI work in production. Not just in the lab. Inference runs through the aic-sdk Python package (also available for Rust, C/C++, Node.js, and WebAssembly).
Setup
- Install the SDK:
pip install aic-sdk - Create a free SDK key on the ai-coustics playground — 30-day trial included, no credit card required.
Usage
# initialize the voice activity detector
import aic_sdk as aic
model = aic.Model.from_file(aic.Model.download("vad-ms-2.1-xxs-16khz", "./models"))
vad = aic.Vad(model, "{aic-sdk-key}", aic.ProcessorConfig.optimal(model)) # get a free SDK key: https://developers.ai-coustics.com/login
# feed mono float32 blocks — the audio is not modified, only analyzed
vad.process(audio_block)
# enjoy real-time, noise-robust turn-taking
speech_detected = vad.get_context().is_speech_detected()
Tuning turn-taking behavior
ctx = vad.get_context()
ctx.set_parameter(aic.VadParameter.Sensitivity, 0.5) # 0.0–1.0 probability threshold
ctx.set_parameter(aic.VadParameter.MinimumSpeechDuration, 0.1) # seconds before "speech" triggers
ctx.set_parameter(aic.VadParameter.SpeechHoldDuration, 0.3) # seconds speech is held after it stops
ctx.raw_vad_probability() exposes the unprocessed per-block probability; ctx.get_prediction_delay() returns the prediction delay in samples for aligning decisions with the input timeline.
Running VAD alongside speech enhancement
Feed the VAD the original input audio, not the enhanced output — run both on the same block:
vad.process(block) # reads the block, does not modify it
enhanced = processor.process(block) # enhances the same original block
Align predictions with your audio timeline via ctx.get_prediction_delay() (delay in samples, like ProcessorContext.get_audio_delay() for enhancement).
Which model should I use?
| Model | Use it for |
|---|---|
VAD Multi Speaker (vad-ms-2.1-xxs-16khz) |
Default choice. Standalone VAD for noisy, multi-speaker, far-field audio — no separate denoiser required. |
VAD Voice Focus (vad-vf-2.0-s-16khz) |
Primary-speaker turn-taking in voice agents. Preferred over chaining Voice Focus enhancement into VAD Multi Speaker. |
Model variants
Weights are hosted on the ai-coustics artifact CDN. aic.Model.download() fetches the latest compatible artifact and verifies its checksum automatically; the direct links below are for manual download.
VAD Multi Speaker (16 kHz)
| SDK model ID | Artifact | SHA-256 (prefix) |
|---|---|---|
vad-ms-2.1-xxs-16khz — latest, recommended |
v7 | 060f15d4e576 |
vad-2.1-xxs-16khz (previous generation) |
v6 | 440cb06bf97a |
quail-vad-2.0-xxs-16khz (legacy ID) |
v5 | 852005789c9b |
VAD Voice Focus (16 kHz)
| SDK model ID | Artifact | SHA-256 (prefix) |
|---|---|---|
vad-vf-2.0-s-16khz — recommended |
v7 | dbbd7995d1cf |
quail-vf-vad-2.0-s-16khz (legacy ID) |
v5 | c50dee9618b8 |
Full SHA-256 checksums for every artifact are listed next to each download at artifacts.ai-coustics.io.
Related ai-coustics models
- 🔊 Speech Enhancement — real-time, ASR-optimized speech enhancement (Quail & Rook)
- 🔍 Audio Insight — predict and diagnose downstream failures in Voice AI stacks
Resources
- 📖 Documentation & model guide: docs.ai-coustics.com
- 🧩 Framework quickstarts: LiveKit, Pipecat
- 📦 Python SDK: pypi.org/project/aic-sdk
- 🔑 Free SDK key: developers.ai-coustics.com
- 🏢 Enterprise licensing: ai-coustics.com