Insanely Fast Whisper
Command-line tool for fast on-device Whisper audio transcription
What it is
A command-line program that transcribes or translates audio files locally using OpenAI's Whisper models through Hugging Face Transformers, with batching, fp16, and optional Flash Attention 2 to cut processing time. It runs on NVIDIA GPUs and Apple Silicon Macs, outputs JSON with chunk or word-level timestamps, and can label speakers via pyannote diarization. People with long recordings, such as podcasts or interviews, use it to get transcripts quickly without sending audio to a cloud service.