playground-transcriber

Simple CLI to transcribe video/audio to plain text using local Whisper from OpenAI.

Install

• Prerequisites: ffmpeg
  • macOS: brew install ffmpeg
  • Ubuntu/Debian: sudo apt-get install -y ffmpeg
• Create virtual environment and install (Linux/OS):
  • python -m transcriber-env .transcriber-env
  • source .transcriber-env/bin/activate
  • pip install -e . #pip install -r requirements.txt

Optional: Faster backend

• pip install faster-whisper
• Use: --backend faster

Optional: Diarization of speakers

• pip install '.[diarize]'
• Get a Hugging Face token:
  • Create a HF account; on the model pages (e.g., pyannote/segmentation and/or pyannote/speaker-diarization), click “Access repository” to accept terms.
  • Copy your token from hf.co/settings/tokens.
• Use: --diarize --hf-token YOUR_HF_TOKEN

Usage

Call from repository main and have your input in a folder called "media" for easiest usage. You can also input the full path if you want to keep the video somewhere else.

• vid2txt input.mp4
• vid2txt /path/to/input.mp4

Give the output a specific name or also save it at another place.

• vid2txt input.mov -o output.txt
• vid2txt input.mov -o /path/to/output.txt

Find some more specifications with: vid2txt --help. For example:

• vid2txt input.mp4 -m medium -l en --device cuda

Optional:

• vid2txt input.mp4 --backend faster
• vid2txt input.mp4 --diarize --hf-token YOUR_HF_TOKEN -o output.txt