sobertranscribe 0.0.2

SoberMind Offline Session Transcriber with Speaker Diarization

SoberMind Session Transcriber

An offline-first, private speech-to-text script utilizing OpenAI's Whisper models for local transcription, with optional PyAnnote.audio integration for multi-speaker diarization (speaker separation).

---

1. System Requirements & Setup

This script runs completely locally on your machine, ensuring absolute confidentiality for your therapy sessions.

Step A: Install FFMPEG

The transcription backend requires ffmpeg to process audio files:

• Windows: Download ffmpeg via chocolatey (choco install ffmpeg) or from the official website, and add its bin directory to your system PATH.
• macOS: brew install ffmpeg
• Linux: sudo apt install ffmpeg

Step B: Install Python Packages

Install the required packages in your Python environment:

pip install openai-whisper torch

---

2. Multi-Speaker Diarization (Optional)

To separate speakers (e.g. distinguishing between Speaker 0 and Speaker 1):

1. Install the diarization dependencies:

   pip install pyannote.audio
   

2. Go to Hugging Face and accept the user agreements for these models (requires creating a free account):
   • pyannote/speaker-diarization-3.1
   • pyannote/segmentation-3.0
3. Generate a User Access Token (Read Permission) on your Hugging Face Settings Page.

---

3. Usage Reference

Standard Transcription (No Speaker Separation)

Runs fully offline immediately:

python transcribe.py path/to/session.mp3

Transcribe with Multi-Speaker Diarization

Splits conversation segments by speaker automatically:

python transcribe.py path/to/session.mp3 --hf-token "YOUR_HF_TOKEN"

Options

• --model: Footprint of model to load (tiny, base, small, medium, large). Defaults to base, which balances speed and accuracy on standard laptops.
• --output: Specify base output name.

Outputs are generated in both:

• .md: A structured Markdown dialogue format.
• .txt: A timestamped plaintext dialogue transcript.

---

4. Web-Based GUI Dashboard

For a premium, interactive editing experience, you can launch the local GUI server:

python gui_server.py [port]

• Default Port: 8080
• Local Address: http://localhost:8080

GUI Features:

1. Drag-and-Drop Form: Easily input your audio target file, Hugging Face Token, and select Whisper model sizes dynamically.
2. Live Console Log: Watch the terminal status updates and model downloads inside a scrollable screen.
3. Dialogue Workspace:
   • Edit transcribed text blocks on the fly.
   • Speaker Renamer: Rename default speaker codes (e.g. SPEAKER_00 to Me, SPEAKER_01 to Dr. Jameson) and instantly replace them across the entire dialogue history.
   • Export Controls: One-click copy formatted Markdown dialogue logs or download local JSON objects.