use-local-whisper

You can transcribe WhatsApp voice notes locally by switching from the Whisper API to a local whisper.cpp workflow. This redirects the transcription pipeline to use a local whisper-cli binary, requiring ffmpeg and a downloaded GGML model to process audio entirely on-device.

Running offline speech-to-text requires installing whisper-cli and ffmpeg, downloading a GGML model file, and ensuring the service PATH includes Homebrew binaries. The WhatsApp transcription code must also be updated to invoke the local whisper-cli binary.

Yes, whisper.cpp can be used for WhatsApp voice transcription in a Docker sandbox. The skill is designed for NanoClaw agent deployments and applies on-device speech-to-text, provided the sandbox PATH is configured to include the necessary Homebrew binaries.

Local whisper-cli transcription fails when the service PATH does not include Homebrew binaries. You must ensure the PATH environment variable for the transcription service is updated to locate the installed whisper-cli and ffmpeg executables correctly.

Local whisper.cpp transcription improves WhatsApp privacy by running fully offline with no external network calls. It eliminates the need for API keys and prevents audio data from leaving the device, making it suitable for private group voice-note transcription.

To set up a whisper.cpp model for offline WhatsApp voice transcription, you need to download a GGML model file. After installing whisper-cli and ffmpeg, update the WhatsApp transcription code to invoke whisper-cli using this local model file.