live-transcribe

Live meeting transcription captures audio using Whisper model speech recognition, supported by PipeWire and PulseAudio for audio routing. It processes audio locally with Python to provide accurate real-time transcripts for business meetings and educational sessions.

Yes, live transcription supports both PipeWire and PulseAudio for audio routing. It offers three distinct audio capture flows tailored to different scenarios, ensuring optimal quality and functionality for your local processing environment.

Local meeting transcription with faster_whisper requires WhisperModel, numpy, pyyaml, and nvidia-cublas-cu12 with nvidia-cudnn-cu12 for GPU acceleration. Ollama and engram dependencies support the real-time Q&A and archiving capabilities.

Yes, real-time Q&A facilitates immediate questions during meetings. You can query the transcript as it is being generated, and end-of-meeting archiving automatically saves the transcript with a summary and relevant notes for future reference.

To start live transcribing a session, use the command `/lt-start`. This initiates the audio capture flow and begins processing the live meeting audio locally through the Whisper model to generate real-time speech recognition transcripts.