herimor/voxtream

VoXtream is a Full-Stream Zero-shot TTS model with Extremely Low Latency and Speaking rate Control

63
/ 100
Established

Built on the Mimi streaming audio codec and classifier-free guidance for dynamic speed adjustment, the model generates speech in real-time chunks with 74ms first-packet latency. Supports multilingual acoustic prompts via prompt text masking, integrates with Gradio for web demos and WebSocket for streaming inference, and achieves 4x faster-than-realtime performance on consumer GPUs with optional CUDA graph compilation.

210 stars and 361 monthly downloads. Available on PyPI.

Maintenance 13 / 25
Adoption 16 / 25
Maturity 18 / 25
Community 16 / 25

How are scores calculated?

Stars

210

Forks

24

Language

Python

License

Apache-2.0

Last pushed

Mar 17, 2026

Monthly downloads

361

Commits (30d)

0

Dependencies

19

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/voice-ai/herimor/voxtream"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.