use-local-whisper

You can achieve local voice transcription by switching to a whisper.cpp implementation that runs entirely on-device. This eliminates external API reliance, removing network calls and API costs for WhatsApp audio transcription.

To set up whisper.cpp transcription on macOS Apple Silicon, install whisper-cpp, apply the voice-transcription skill first, and place a GGML model file in the data/models directory.

On-device whisper transcription currently supports WhatsApp channels. Extending to other channels requires implementing channel-specific audio download logic to fetch the voice messages before transcription.

Local voice transcription with whisper.cpp does not require an internet connection. The process runs entirely on-device, offering a privacy-centric, low-latency transcription solution without network calls.

Your local whisper transcription requires a GGML model file because whisper.cpp uses this specific format to run inference. You must have the model file available in the data/models directory to process audio.