# Whisper Transcription

> Speech-to-text workflows using Whisper locally or via the OpenAI transcription API. Use when transcribing audio, choosing a Whisper runtime, or producing text, SRT, or JSON transcripts.

- Skill: `s3yed/whisper-transcription` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add s3yed/whisper-transcription`
- Raw SKILL.md: https://api.skillmd.com/api/skills/s3yed/whisper-transcription/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Integrations & APIs
- License: MIT
- Author: S3YED (https://skillmd.com/u/s3yed)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/s3yed/whisper-transcription

---


# Whisper Transcription

## Overview
Use this umbrella for audio transcription tasks regardless of whether the runtime is local Whisper or the OpenAI API.

## Choose a Path
- **Local CLI:** when you want offline transcription and already have the Whisper binary.
- **OpenAI API:** when you want hosted transcription, JSON output, or proxy-compatible API access.

## Workflow
1. Identify the source audio format and desired output (txt, srt, json).
2. Pick local or API based on latency, privacy, and environment constraints.
3. Set model/language/prompt hints only when they improve the transcript.
4. Verify the transcript against the audio length and obvious speaker names.

## References
- `references/path-selection.md` for runtime choice and output-format notes.

## Pitfalls
- Don’t keep separate skills for local and API Whisper when the task class is identical.
- Don’t overfit model choice; prefer the simplest path that meets quality needs.
- Don’t forget to specify output format and destination.