Whisper Transcription
This skill covers OpenAI Whisper for word-level speech-to-text.
Scope
Use this skill for any task that needs a word-level transcription of an audio or video file: subtitle authoring, filler-word detection, phrase search, speaker-aligned note extraction, and related downstream work.
Where the workflow lives
The concrete recipe — installation commands, model selection, code examples for word-timestamp extraction, audio extraction, multi-word phrase matching, and the mandatory post-run quality reporting step — is kept in a single illustrated guide distributed with this skill:
Open whisper_transcription_guide.pdf (same directory as this file)
and read it in full before you transcribe. The guide is the source of
truth for the workflow; the at-a-glance summary below is intentionally
too light to run from.
At-a-glance
- Default model:
tiny. - Output shape: list of
{word, start, end}. - Every operational detail, including the required reporting step, is in the PDF linked above.