Starstruk Logo MarkStarstruk

v0.2.19

mikeqzy/3dgs-avatar-releasev0.2.19Feb 23, 2026by QuentinFuxa

AI Summary

Adds a new 'Voxtral' backend for speech recognition and introduces a comprehensive offline benchmarking framework.

Key Highlights

  • Added native MLX backend for Apple Silicon (Voxtral-mlx).
  • Added HuggingFace Transformers backend for Linux/GPU.
  • Introduced offline benchmark harness to compute WER and RTF metrics.
  • Fixed silence double-counting and RTF inflation in metrics.

New Features

  • Voxtral-MLX backend (Apple Silicon).
  • Voxtral HuggingFace backend (Linux/GPU).
  • Offline benchmark harness for backends.

Full Release Notes

## Voxtral backend & benchmarks

### New: Voxtral backend

- **voxtral-mlx**: Native MLX backend for Apple Silicon. Runs at 0.18-0.32x real-time, handles 100+ languages with automatic language detection. No extra dependencies needed on macOS.
- **voxtral (HF)**: HuggingFace transformers backend for Linux/GPU. Requires `pip install transformers torch`.

### Benchmarks

New offline benchmark harness (`test_backend_offline.py --benchmark`) that runs all installed backends and computes WER, RTF, and timestamp accuracy against ground truth transcripts. Results exportable as JSON.

Full benchmark report in [BENCHMARK.md](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/BENCHMARK.md) with tables, charts, and recommendations for every backend/policy/model combination.

### Bug fixes

- Fixed silence double-counting in the audio processor
- Fixed median calculation for even-length lists in timestamp accuracy
- Fixed RTF inflation in metrics collector (was using wall-clock time instead of ASR processing time)