Lyrics Transcription Skill

Transcribe audio files to timestamped lyrics (LRC/SRT/JSON) via OpenAI Whisper or ElevenLabs Scribe API.

API Key Setup Guide

Before transcribing, you MUST check whether the user's API key is configured. Run the following command to check:

CODEBLOCK0

This command only reports whether the active provider's API key is set or empty — it does NOT print the actual key value. NEVER read or display the user's API key content. Do not use config --get on key fields or read config.json directly. The config --list command is safe — it automatically masks API keys as *** in output.

If the command reports the key is empty, you MUST stop and guide the user to configure it before proceeding. Do NOT attempt transcription without a valid key — it will fail.

Use AskUserQuestion to ask the user to provide their API key, with the following options and guidance:

1. 1. Tell the user which provider is currently active (openai or elevenlabs) and that its API key is not configured. Explain that transcription cannot proceed without it.
2. Provide clear instructions on where to obtain a key:

- OpenAI: Get an API key at https://platform.openai.com/api-keys — requires an OpenAI account with billing enabled. The Whisper API costs ~$0.006/min. - ElevenLabs: Get an API key at https://elevenlabs.io/app/settings/api-keys — requires an ElevenLabs account. Free tier includes limited credits.

1. 3. Also offer the option to switch to the other provider if they already have a key for it.
2. Once the user provides the key, configure it using:

   cd "{project_root}/{.claude or .codex}/skills/acestep-lyrics-transcription/" && bash ./scripts/acestep-lyrics-transcription.sh config --set <provider>.api_key <KEY>
   

1. 5. If the user wants to switch providers, also run:

   cd "{project_root}/{.claude or .codex}/skills/acestep-lyrics-transcription/" && bash ./scripts/acestep-lyrics-transcription.sh config --set provider <provider_name>
   

1. 6. After configuring, re-run config --check-key to verify the key is set before proceeding.

If the API key is already configured, proceed directly to transcription without asking.

Quick Start

CODEBLOCK3

Prerequisites

• - curl, jq, python3 (or python)
• An API key for OpenAI or ElevenLabs

Script Usage

CODEBLOCK4

Post-Transcription Lyrics Correction (MANDATORY)

CRITICAL: After transcription, you MUST manually correct the LRC file before using it for MV rendering. Transcription models frequently produce errors on sung lyrics:

• - Proper nouns: "ACE-Step" → "AC step", "Spotify" → "spot a fly"
• Similar-sounding words: "arrives" → "eyes", "open source" → "open sores"
• Merged/split words: "lighting up" → "lightin' nup"

Correction Workflow

1. 1. Read the transcribed LRC file using the Read tool
2. Read the original lyrics from the ACE-Step output JSON file
3. Use original lyrics as a whole reference: Do NOT attempt line-by-line alignment — transcription often splits, merges, or reorders lines differently from the original. Instead, read the original lyrics in full to understand the correct wording, then scan each LRC line and fix any misrecognized words based on your knowledge of what the original lyrics say.
4. Fix transcription errors: Replace misrecognized words with the correct original words, keeping the timestamps intact
5. Write the corrected LRC back using the Write tool

What to Correct

• - Replace misrecognized words with their correct original versions
• Keep all [MM:SS.cc] timestamps exactly as-is (timestamps from transcription are accurate)
• Do NOT add structure tags like [Verse] or [Chorus] — the LRC should only have timestamped text lines

Example

Transcribed (wrong):
CODEBLOCK5

Original lyrics reference:
CODEBLOCK6

Corrected (right):
CODEBLOCK7

Configuration

Config file: INLINECODE9

CODEBLOCK8

Option	Default	Description
INLINECODE10	INLINECODE11	Active provider: openai or INLINECODE13
INLINECODE14

lrc | Default output: lrc, srt, or json |
| openai.api_key | "" | OpenAI API key |
| openai.api_url | https://api.openai.com/v1 | OpenAI API base URL |
| openai.model | whisper-1 | OpenAI model (whisper-1 for word timestamps) |
| elevenlabs.api_key | "" | ElevenLabs API key |
| elevenlabs.api_url | https://api.elevenlabs.io/v1 | ElevenLabs API base URL |
| elevenlabs.model | scribe_v2 | ElevenLabs model |

Provider Notes

Provider	Model	Word Timestamps	Pricing
OpenAI	whisper-1	Yes (segment + word)	$0.006/min
ElevenLabs

scribe_v2 | Yes (word-level) | Varies by plan |

• - OpenAI whisper-1 is the only OpenAI model supporting word-level timestamps
• ElevenLabs scribe_v2 returns word-level timestamps with type filtering
• Both support multilingual transcription

Examples

CODEBLOCK9