GetTranscript

A local-first audio transcription toolkit — from YouTube video to polished transcript in three commands.

Website · GitHub

A local-first audio transcription toolkit that converts YouTube videos or audio files into publication-ready Chinese transcripts. The pipeline runs in three stages: audio extraction, GPU-accelerated speech recognition, and LLM-powered punctuation.

Pipeline:

  • YouTube auto-download via yt-dlp to extract highest quality WAV audio
  • Local Whisper large-v3 transcription using faster-whisper with GPU batch inference and checkpoint resume
  • LLM punctuation via DeepSeek API for intelligent sentence segmentation, punctuation, and paragraph formatting

Highlights:

  • All speech recognition runs locally on GPU — audio never leaves your machine
  • Batch processing with checkpoint support for pause/resume
  • Supports float16, int8, and CPU fallback for flexible hardware

Technologies: Python, faster-whisper, yt-dlp, CUDA, DeepSeek API, FFmpeg