Changelog

Changelog

v1.0.0

Initial public release.

Features

• Batched decoding — 3-5x faster transcription via parallel segment processing on Apple Silicon
• LightningWhisperMLX — simple Python API for common use cases
• Full transcribe() API — all Whisper options exposed
• CLI (vayu command) — full-featured command-line interface with 30+ options
• All Whisper models — tiny through large-v3, turbo, and distil variants
• Quantization — 4-bit and 8-bit quantized models for reduced memory
• Word-level timestamps — via cross-attention + DTW alignment
• 5 output formats — txt, srt, vtt, tsv, json
• 99 languages — full multilingual support with auto-detection
• Translation — translate any language to English
• Speculative decoding (experimental) — additional 2-3x speedup with draft/target model pairs
• VAD processing — voice activity detection for silence skipping
• Parallel chunk transcription — process long audio with overlapping chunks
• Security — model path validation to prevent path traversal attacks
• Temperature fallback — automatic quality recovery via temperature escalation
• Subtitle formatting — word highlighting, line width/count controls for SRT/VTT
• Batch file processing — transcribe multiple files with error tolerance
• stdin support — pipe audio from other processes

Requirements

• macOS with Apple Silicon (M1/M2/M3/M4)
• Python 3.10+
• MLX 0.11+

---

For the latest changes, see the GitHub Releases page.