TrevorS/voxtral-mini-realtime-rs

Streaming speech recognition running natively and in the browser. A pure Rust implementation of Mistral's Voxtral Mini 4B Realtime model using the Burn ML framework.

52
/ 100
Established

Supports both ASR and TTS with dual inference paths optimized for native (BF16/Q4 GGUF) and browser (WASM + WebGPU) execution, achieving real-time synthesis at 0.97 RTF on Q4 with custom WGSL kernels and fused operations. Solves five critical WASM constraints including 2GB allocation limits and 1.5GB embedding tables via sharded loading and GPU-resident quantized embeddings. Integrates with HuggingFace Hub for model distribution, Burn ML framework for compute, and cubecl-wgpu for WebGPU shader dispatch.

710 stars. Actively maintained with 25 commits in the last 30 days.

No Package No Dependents
Maintenance 20 / 25
Adoption 10 / 25
Maturity 9 / 25
Community 13 / 25

How are scores calculated?

Stars

710

Forks

33

Language

Rust

License

Apache-2.0

Last pushed

Feb 12, 2026

Commits (30d)

25

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/voice-ai/TrevorS/voxtral-mini-realtime-rs"

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