transcribe-maker

<objective> Speech-to-text utility. Take audio or video file → output subtitle file (SRT / VTT) or transcript (JSON / text). Distinct from `voiceover-maker`: - voiceover-maker generates SPEECH from text. This goes the OPPOSITE direction: speech → text. - Both wrap an OpenAI audio API endpoint. Distinct from `subtitle-burner`: - subtitle-burner takes an existing subtitle file and burns it into video. THIS skill GENERATES the subtitle file. Chain them: `transcribe-maker` → `subtitle-burner`. This skill does NOT: - Translate the transcript (Whisper supports it via separate endpoint; not exposed in v1 — use OpenAI's `/v1/audio/translations` directly if needed) - Identify speakers / diarize (Whisper doesn't natively diarize — use deepgram or assemblyai for that) - Process files >25 MB (Whisper API limit — split with ffmpeg first) - Run locally (uses OpenAI API; for offline use whisper.cpp directly) - Edit / clean up transcript (raw Whisper output may have minor errors — manual cleanup needed for production captions) </objective> ## ROLE Read audio/video input → call OpenAI Whisper API with the requested format → save subtitle / transcript file. ## PIPELINE 1. **Resolve input**: - `--input <path>` — audio (.mp3, .wav, .m4a) or video (.mp4, .mov, .webm) (required) 2. **Pick output format**: - `--format srt` (default) — for `subtitle-burner` consumption - `--format vtt` — WebVTT (YouTube / HTML5 `<track>`) - `--format text` — plain transcript, no timestamps - `...