pyannoteAI Live-1 is now generally available. This streaming diarization model is designed specifically for live audio pipelines. Accuracy, latency, and noise robustness are optimized for production streaming conditions, not retrofitted from batch processing.

Capabilities:

• Delivers speaker diarization on live audio, with under 300ms latency

• Handles background noise, overlapping speech, and variable audio quality

• Structured metadata output: speaker turns, labels, timestamps, updated continuously

• Scalable infrastructure supporting parallel streams without latency spikes

Benchmark results:

Integration:
Compatible with any STT pipeline via WebSocket API. You can add Live-1, the streaming diarization model, alongside your existing Voice AI Pipeline.

Pricing:

• Developer Plan: €0.198/hour

• Starter Plan: €0.170/hour

• Enterprise Plan: Contact us for details

👉 Explore technical documentation
📝 Access to the tutorial to build a streaming diarized transcription tool
📚 Read about how we built our model in our blog post