v0.2.19

hatchet-dev/icepickv0.2.19Feb 23, 2026by QuentinFuxa

AI Summary

Adds support for the Voxtral backend, including native MLX support for Apple Silicon and HuggingFace transformers for Linux/GPU, alongside a new offline benchmarking harness.

Key Highlights

Native MLX backend (voxtral-mlx) for Apple Silicon with 0.18-0.32x real-time speed
HuggingFace transformers backend (voxtral) for Linux/GPU
New offline benchmark harness computing WER, RTF, and timestamp accuracy
Bug fixes for RTF inflation and silence double-counting

New Features

Voxtral backends (MLX and HF)
Offline benchmarking harness
WER and RTF metric fixes

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)