v0.2.19

SimonGiebenhain/MonoNPHMv0.2.19Feb 23, 2026by QuentinFuxa

AI Summary

Introduces a new Voxtral backend for Apple Silicon and Linux/GPU, adds an offline benchmarking harness for evaluating backends, and fixes several audio processing and metrics bugs.

Key Highlights

Native MLX backend for macOS (voxtral-mlx) and HuggingFace backend for Linux/GPU
New offline benchmark harness with JSON export for WER, RTF, and timestamp accuracy
Fixed silence double-counting in the audio processor
Fixed RTF inflation in the metrics collector

New Features

Native MLX backend (voxtral-mlx) for Apple Silicon
HuggingFace transformers backend (voxtral HF) for Linux/GPU
Offline benchmark tool (test_backend_offline.py)
Comprehensive benchmark report in BENCHMARK.md

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)