use-local-whisper

Local transcription of WhatsApp voice messages is achieved by switching to whisper.cpp with a modified transcription.ts file, requiring no API key or network connection. You simply install whisper-cli and ffmpeg, then place a GGML model in the data/models directory.

Yes, on-device audio processing with whisper.cpp is specifically supported on Apple Silicon macOS. It uses the whisper-cli executable to run transcription locally without sending audio data to the cloud.

To set up local transcription, you need to install whisper-cpp and ffmpeg on your system. Additionally, you must download a GGML model and place it in the data/models directory before running the transcription process.

Local whisper transcription currently supports WhatsApp directly, but it can be extended to other channels. To support additional platforms, you must implement channel-specific audio download logic to feed audio files into the transcription pipeline.

Local transcription fails if whisper-cli is not installed, ffmpeg is not available in your system path, or the required GGML model is missing from the data/models directory. Ensure your transcription.ts file is modified to call whisper-cli instead of the OpenAI API.