transcribe
Summary
Transcribe audio files to text with optional speaker diarization and known-speaker hints.
- Supports fast text transcription via
gpt-4o-mini-transcribeand speaker-labeled diarization viagpt-4o-transcribe-diarize - Accepts multiple audio formats and optional known-speaker references (up to 4 speakers) to improve diarization accuracy
- Outputs as plain text, JSON, or diarized JSON with configurable output directories to prevent overwrites
- Requires
OPENAI_API_KEYenvironment variable; uses bundled Python CLI for deterministic, repeatable transcription runs
SKILL.md
Audio Transcribe
Transcribe audio using OpenAI, with optional speaker diarization when requested. Prefer the bundled CLI for deterministic, repeatable runs.
Workflow
- Collect inputs: audio file path(s), desired response format (text/json/diarized_json), optional language hint, and any known speaker references.
- Verify
OPENAI_API_KEYis set. If missing, ask the user to set it locally (do not ask them to paste the key). - Run the bundled
transcribe_diarize.pyCLI with sensible defaults (fast text transcription). - Validate the output: transcription quality, speaker labels, and segment boundaries; iterate with a single targeted change if needed.
- Save outputs under
output/transcribe/when working in this repo.
Decision rules
- Default to
gpt-4o-mini-transcribewith--response-format textfor fast transcription. - If the user wants speaker labels or diarization, use
--model gpt-4o-transcribe-diarize --response-format diarized_json. - If audio is longer than ~30 seconds, keep
--chunking-strategy auto. - Prompting is not supported for
gpt-4o-transcribe-diarize.
Output conventions
- Use
output/transcribe/<job-id>/for evaluation runs. - Use
--out-dirfor multiple files to avoid overwriting.
Dependencies (install if missing)
Prefer uv for dependency management.
uv pip install openai
If uv is unavailable:
python3 -m pip install openai
Environment
OPENAI_API_KEYmust be set for live API calls.- If the key is missing, instruct the user to create one in the OpenAI platform UI and export it in their shell.
- Never ask the user to paste the full key in chat.
Skill path (set once)
export CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
export TRANSCRIBE_CLI="$CODEX_HOME/skills/transcribe/scripts/transcribe_diarize.py"
User-scoped skills install under $CODEX_HOME/skills (default: ~/.codex/skills).
CLI quick start
Single file (fast text default):
python3 "$TRANSCRIBE_CLI" \
path/to/audio.wav \
--out transcript.txt
Diarization with known speakers (up to 4):
python3 "$TRANSCRIBE_CLI" \
meeting.m4a \
--model gpt-4o-transcribe-diarize \
--known-speaker "Alice=refs/alice.wav" \
--known-speaker "Bob=refs/bob.wav" \
--response-format diarized_json \
--out-dir output/transcribe/meeting
Plain text output (explicit):
python3 "$TRANSCRIBE_CLI" \
interview.mp3 \
--response-format text \
--out interview.txt
Reference map
references/api.md: supported formats, limits, response formats, and known-speaker notes.
Weekly Installs
562
Repository
openai/skillsGitHub Stars
15.6K
First Seen
Feb 1, 2026
Security Audits
Installed on
codex504
opencode485
gemini-cli476
github-copilot463
cursor457
kimi-cli454