Local audio transcription

Run zsh scripts/transcribe_audio.sh "audio-path" from this skill directory. Default to Italian; use --language en when the user specifies English. Respect another supported language when explicitly requested.

The helper requires ffmpeg, whisper-cli and a local GGML model. It uses $WHISPER_MODEL_PATH when present; --model "model-path" overrides it.

The helper creates YYYY-MM-DD_HH-MM_file-name/ beside the source, containing a copy of the original, a mono 16 kHz PCM WAV and *_transcript.txt. Keep generated files there even for WAV input. Preserve the source.

Verify successful completion and a usable transcript, then return the transcript and output-folder links. Produce subtitles only when requested.

Local configuration

Run the helper with zsh or directly via its shebang, not bash. Model precedence is --model, then WHISPER_MODEL_PATH, then a model in WHISPER_MODEL_DIR (default $HOME/.local/share/whisper/models). Prefer ggml-medium.bin, then an available ggml-*.bin. Do not assume a particular user home or model installation.