| name | whisper-audio-transcriber |
| description | Use when you need to transcribe a local audio file with whisper-cpp. The skill creates a timestamped folder next to the source audio, copies the original file there, converts it to a Whisper-friendly WAV, and writes a plain-text transcript. Default language is Italian; optionally force English. |
Whisper Audio Transcriber
Use this skill for one-off local transcriptions with whisper-cli and ffmpeg.
Default language is Italian. Use English only when the source audio is clearly in English.
Workflow
- Confirm the input audio path.
- Run
scripts/transcribe_audio.sh with the audio path.
- By default, do not pass a language flag, so the script uses Italian.
- If the user says the audio is in English, pass
--language en.
- Return the output folder path and the transcript path.
Command
Italian by default:
/Users/mameli/.codex/skills/whisper-audio-transcriber/scripts/transcribe_audio.sh /absolute/path/to/file.m4a
Force English:
/Users/mameli/.codex/skills/whisper-audio-transcriber/scripts/transcribe_audio.sh --language en /absolute/path/to/file.m4a
Optional model override:
/Users/mameli/.codex/skills/whisper-audio-transcriber/scripts/transcribe_audio.sh --model /Users/mameli/Ai_models/ggml-medium.bin /absolute/path/to/file.m4a
Output Layout
The script creates a folder next to the source file with this naming pattern:
YYYY-MM-DD_HH-MM_file-name/
Inside it, write:
- the original audio file
*_whisper.wav converted to mono, 16 kHz, pcm_s16le
*_transcript.txt as plain text
Requirements
ffmpeg
whisper-cli
- a local GGML model, defaulting to
/Users/mameli/Ai_models/ggml-medium.bin when present
If the default model is missing, pass --model.
Notes
- Keep the transcript as plain text only. Do not request subtitle output unless the user asks.
- If the input audio is already WAV, still place all generated files inside the timestamped folder.
- If language is not specified by the user, use Italian.