SynapseMusicV12-Transcriber

<a href="https://arxiv.org/abs/2602.00744">Tech Report</a>

ACE-Step Transcriber

Description

ACE-Step Transcriber is the annotation model used by ACE-Step v1.5 for training data labeling. It is a powerful multilingual audio transcription model capable of transcribing both speech and singing voice with high accuracy.

Key Features

• 🌍 50+ Languages Support - Covers major world languages and regional dialects
• 🎤 Speech Transcription - Accurately transcribes spoken content
• 🎵 Singing Voice Transcription - Specialized in lyrics transcription with musical structure annotations
• 🏷️ Structure Annotation - Automatically identifies song sections (verse, chorus, bridge, etc.)

Usage

The usage is the same as Qwen2.5 Omni-7B.

Prompt Format

Use the following prompt to transcribe audio:

*Task* Transcribe this audio in detail
<audio>

Output Format

The model outputs structured content in the following format:

# Languages
<language_code>

# Lyrics
[Section Tag - Optional Instrument]

<transcribed content>
...

Example Output

# Languages
en

# Lyrics
[Intro - Acoustic Guitar]

[Verse 1]
Walking down the empty street tonight
Stars are shining oh so bright
...

[Chorus]
This is where we belong
Singing our favorite song
...

Supported Section Tags

• [Intro], [Outro]
• [Verse 1], [Verse 2], etc.
• [Chorus], [Pre-Chorus], [Post-Chorus]
• [Bridge]
• [Guitar Interlude], [Instrumental]
• [Spoken]

Supported Languages (50+)

The model supports transcription in over 50 languages, including but not limited to:

Use Cases

• Music Production - Transcribe reference tracks for lyrics extraction
• Dataset Creation - Generate high-quality labeled data for music AI models
• Accessibility - Create subtitles and captions for audio content
• Music Analysis - Extract structural information from songs