Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Voxtral ASR & TTS running natively and in the browser. A Rust implementation of Mistral's Voxtral mini realtime ASR / TTS using the Burn ML framework
| Date | Stars |
|---|---|
| 2026-07-24 | 810 |
| 2026-07-25 | 810 |
| 2026-07-28 | 810 |
| 2026-07-30 | 810 |
| 2026-08-06 | 810 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Voxtral Mini 4B Realtime (Rust) [](https://huggingface.co/TrevorJS/voxtral-mini-realtime-gguf) [](https://huggingface.co/TrevorJS/voxtral-tts-q4-gguf) [](https://huggingface.co/spaces/TrevorJS/voxtral-mini-realtime) [](https://huggingface.co/spaces/TrevorJS/voxtral-4b-tts) Streaming speech recognition and text-to-speech running natively and in the browser. A pure Rust implementation of Mistral's [Voxtral Mini 4B Realtime](https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-2602) (ASR) and [Voxtral 4B TTS](https://huggingface.co/mistralai/Voxtral-4B-TTS-2603) models using the [Burn](https://burn.dev) ML framework. ## Benchmarks NVIDIA DGX Spark (GB10, LPDDR5x). ### ASR (Speech Recognition) 16s test audio, 3-run average: | Path | Encode | Decode | Total | RTF | Tok/s | Memory | |------|--------|--------|-------|-----|-------|--------| | **Q4 GGUF native** | 1021 ms | 5578 ms | 6629 ms | **0.416** | **19.4** | 703 MB | | BF16 native | 887 ms | 23689 ms | 24607 ms | 1.543 | 4.6 | 9.2 GB | | Q4 GGUF WASM | — | — | ~225 s | ~14.1 | ~0.5 | (browser) | - **8.49% WER** on FLEURS English (647 utterances), vs. Mistral's reported 4.90% at f32 ### TTS (Text-to-Speech) "The quick brown fox jumps over the lazy dog" (9 tokens), casual_female voice: | Path | Euler Steps | Gen Time | Audio | RTF | Model Size | |------|-------------|----------|-------|-----|------------| | **Q4 GGUF native** | 3 | 3.7s | 3.84s | **0.97** | 2.67 GB | | Q4 GGUF native | 4 | 5.0s | 4.96s | 1.01 | 2.67 GB | | BF16 native | 3 | 10.4s | 2.72s | 3.82 | ~8 GB | | BF16 native | 8 | 20.6s | 2.96s | 6.97 | ~8 GB | | Q4 GGUF WASM | 8 | 367s | 3.52s | 104 | 2.67 GB | - **RTF** < 1.0 means faster-than-real-time synthesis - Q4 at 3 Euler steps achieves **real-time** with perfect Whisper large-v3 transcription - Optimizations: batched CFG (2× → batch=2), fused QKV+gate/up projections, pre-allocated KV cache - Q4 model load: 3.9s native, 9.2s WASM (including shard download over localhost) - 20 preset voices across 9 languages. Use `--euler-steps` to tune speed/quality tradeoff ### Architecture Notes - Custom WGSL compute shaders with vectorized u32 reads and vec4 dot products - Dual-path kernel dispatch: shared-memory tiled kernel for single-token decode, naive kernel for multi-row encode/prefill - Q4 GGUF (2.5 GB ASR, 2.67 GB TTS) runs entirely client-side in a browser tab via WASM + WebGPU Try the demos: [ASR (speech-to-text)](https://huggingface.co/spaces/TrevorJS/voxtral-mini-realtime) | [TTS (text-to-speech)](https://huggingface.co/spaces/TrevorJS/voxtral-4b-tts) ## Quick Start ### Native CLI ```bash # Download ASR model weights (~9 GB BF16 or ~2.5 GB Q4) uv run --with huggingface_hub \ hf download mistralai/Voxtral-Mini-4B-Realtime-2602 --local-dir models/voxtral uv run --with huggingface_hub \ hf download TrevorJS/voxtral-mini-realtime-gguf --local-dir models/ # Transcribe audio (BF16 or Q4) cargo run --release --features "wgpu,cli,hub" --bin voxtral -- \ transcribe --audio audio.wav --model models/voxtral cargo run --release --features "wgpu,cli,hub" --bin voxtral -- \ transcribe --audio audio.wav --gguf models/voxtral-q4.gguf ``` ### Browser Demo ```bash # Build WASM package wasm-pack build --target web --no-default-features --features wasm # Generate self-signed cert (WebGPU requires secure context) openssl req -x509 -newkey ec -pkeyopt ec_paramgen_curve:prime256v1 \ -keyout /tmp/voxtral-key.pem -out /tmp/voxtral-cert.pem \ -days 7 -nodes -subj "/CN=localhost" # Start dev server bun serve.mjs ``` Open `https://localhost:8443`, accept the certificate, and click **Load from Server** to download the model shards. Record from your m
Excerpt of 10,017 characters
Read on GitHub
52
Johnny Shields · TableCheck · Japan
2
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:ed86de68ef0441e9, topic:tts, topic:asr, readme:speech recognition
matched fp:ed86de68ef0441e9, topic:mistral, readme:model weights