audio-transcription

Local audio transcription generates timestamped segments entirely offline using whisper.cpp. This privacy-preserving approach processes audio intake without external network calls, making it suitable for secure clinical documentation workflows.

Yes, offline transcription with whisper.cpp supports optional speaker diarization. This feature separates different speakers in the audio intake, producing labeled, timestamped transcript segments for multi-speaker clinical documentation.

Local transcription supports common audio formats like WAV for offline processing. You provide a local audio file to the workflow, which decodes the input and generates a timestamped transcript.json artifact.

Offline transcription post-processing includes number and date normalization alongside punctuation restoration. These deterministic steps refine the raw whisper.cpp output, delivering clean, formatted text ready for clinical artifact persistence.

Model selection involves choosing a whisper.cpp model size that balances accuracy and local processing speed. The skill generates a model_selection.md artifact to guide your configuration for optimal offline transcription results.

Privacy-preserving local transcription relies entirely on local compute resources, meaning processing speed depends on your hardware. The workflow includes an error_handling.md artifact to address known offline decoding and diarization constraints.