sigh

Seamless Voice Interactions with LLMs

Key Features:

• Unlimited Real-time Transcription: Continuously capture audio directly from your microphone.
• Customizable Wake Word: Choose a wake word or phrase to trigger transcription mode.
• Automatic Speech Termination: Detects when you've finished speaking, with an option for manual control.

Note: This repository is under active development. Contributions are welcome!

Demo:

Setup:

set OPENAI_API_KEY=sk-...

git clone https://github.com/eryk-mazus/sigh.git
cd sigh
pip install -e .

# run:
python ./sigh/main.py --help

# run without wake word detection (by default):
python ./sigh/main.py

# run with wake phrase detection:
python ./sigh/main.py --detect_wake_phrase=True --wake_phrase="""Hey GPT"""

Backlog:

Near-term:

• Add automatic transcription stopping
• Better GPT responses (system prompt, chat mode, sliding memory buffer)
• Talk with local models, e.g. llama2, mistral, etc.
• Improve code coherence and composition (refactoring)

Medium-term:

• Add second mode: parallel transcription and LLM commentary
• Docker

Contributing:

Issues, new ideas, suggestions, and PRs are all welcome!