Home

Vayu Wiki

Vayu (وایو) is the fastest Whisper speech-to-text implementation optimized for Apple Silicon Macs. It combines MLX Whisper with Lightning Whisper MLX batched decoding to deliver 3-5x faster transcription than standard Whisper on M1/M2/M3/M4 chips.

Named after the ancient Persian god of wind — the swiftest force in nature.

Getting Started

• Installation — Setup instructions and requirements
• Quick Start — Get transcribing in minutes
• Troubleshooting — Common issues and FAQ

Reference

• Python API — Full Python API reference
• CLI Reference — Command-line interface documentation
• Models — Available models, quantization, and batch size recommendations
• Output Formats — Supported output formats (txt, srt, vtt, tsv, json)

Guides

• Multilingual Guide — 99 languages, RTL, CJK, translation
• Word Timestamps Guide — Precise word timings, subtitle workflows
• Speculative Decoding Guide — Experimental 2-3x additional speedup
• Examples and Use Cases — Real-world recipes and workflows
• Integration Recipes — FastAPI, Flask, FFmpeg, batch scripts, cron jobs
• Comparison — Vayu vs whisper.cpp vs faster-whisper vs OpenAI Whisper

Project

• Architecture — Internal architecture and design
• Performance — Optimizations and tuning tips
• Changelog — Version history
• Contributing — How to contribute to Vayu

Key Features

• 3-5x faster than standard Whisper via batched decoding on Apple Silicon
• One-line install — uv pip install vayu-whisper or pip install vayu-whisper
• All Whisper models — tiny through large-v3, plus turbo and distil variants
• Multiple output formats — txt, vtt, srt, tsv, json
• Word-level timestamps — precise word timings for subtitles and alignment
• Low memory options — 4-bit and 8-bit quantization
• Simple Python API + CLI

Links

• PyPI Package
• GitHub Repository
• Issue Tracker