scribe

Utility to record and transcribe meetings. Uses Sox for cross-platform audio recording. Supports optional speaker diarization (who spoke when) and per-segment sentiment analysis.

Installation

# Install Python dependencies
pdm install

# Install Sox (required for audio recording)
# macOS:
brew install sox

# Linux (Ubuntu/Debian):
sudo apt install sox

# Linux (RedHat/CentOS):
sudo yum install sox

Speaker diarization (optional)

Diarization is off by default; pass --diarization to enable it. You can use it in two ways:

Option 1: No account (free)
If you do not set HF_TOKEN, scribe uses Resemblyzer to diarize: it embeds each transcript segment and clusters by voice. No Hugging Face account or token is required. Install the extra deps (included in pdm install): Resemblyzer, scikit-learn.

Option 2: Pyannote (best quality)
For higher-quality diarization, use Hugging Face and pyannote:

1. Create a token at hf.co/settings/tokens.
2. Accept the user conditions for:
   • pyannote/speaker-diarization-3.1
   • pyannote/segmentation-3.0
3. Set the token when running:

   export HF_TOKEN=your_token_here
   pdm run meet

   Or use HUGGING_FACE_HUB_TOKEN instead of HF_TOKEN.

To enable diarization, pass the flag:

pdm run meet --diarization

Usage

# Record and transcribe only (default)
pdm run meet

# Enable diarization and/or sentiment
pdm run meet --diarization
pdm run meet --sentiment
pdm run meet --diarization --sentiment

# Record audio only
pdm run record my-recording                    # With timestamp
pdm run record my-recording --no-timestamp     # Without timestamp

Output

Creates a folder in records/ with:

• Audio file (WAV → MP3)
• Transcription (TXT) with timestamps. With diarization and sentiment enabled, each line looks like:

  [0.00s -> 5.20s] Speaker_0 (positive): Hello everyone, thanks for joining.
  [5.20s -> 12.40s] Speaker_1 (neutral): Let's go through the agenda.
  

  With both disabled, the format is backwards compatible: [start->end] text.