A local-first audio transcription toolkit. Download, transcribe, and polish — from YouTube video to publication-ready transcript in three commands.
Any video or audio file
GPU-accelerated STT
DeepSeek API formatting
Clean, readable text
Pass a video or playlist URL. The tool uses yt-dlp to extract the highest quality WAV audio automatically.
Runs OpenAI's Whisper large-v3 locally via faster-whisper. Supports GPU batch inference, checkpoint resume, and float16/int8 precision modes.
Raw speech-to-text output is messy. The LLM post-processor adds punctuation, paragraph breaks, and natural formatting to produce publication-ready transcripts.
All speech recognition runs on your own GPU. Only the text formatting step optionally calls an external API — your audio never leaves your machine.
Process entire folders of audio files. Checkpoint support means you can pause and resume without losing progress.
Best with 8GB+ VRAM GPU. Falls back to int8 for smaller GPUs, or CPU mode if no NVIDIA card is available.