Audio Structuring Skill

Purpose

Convert a meeting audio file into a dialogue-style transcript txt file for downstream skills.

Input

• audio_path: path to a single audio file
• output_dir: directory to save the transcript
• optional transcription_language: language code such as en or zh

Output

• meeting_transcript.txt

Behavior

• Reuse WhisperX as the transcription backend.
• Reuse local offline model directories instead of downloading models from Hugging Face at runtime.
• Enable diarization so the txt contains speaker labels whenever available.
• Automatically convert Traditional Chinese output to Simplified Chinese when language is set to zh, zh-CN, or zh-TW.
• Do not implement custom ASR, alignment, or diarization logic.
• Do not generate summary, action items, reports, or any extra files.
• This skill only produces transcript txt for downstream skills.

Local model requirements

The repository is expected to provide these local model directories under models/:

• models/faster-whisper-large-v2
• models/speaker-diarization-community-1

Run

bash scripts/run.sh <audio_path> <output_dir> [language]