How andyanttle/ace-step-captioner-transcriber works
ACE-step Captioner & Transcriber is a Gradio application for captioning music with Qwen Omni or transcribing it with Whisper. The user selects files or a folder, chooses a model and preprocessing options, runs recognition, then views and saves results. Processing includes audio loading, optional stem separation and noise reduction, duration trimming, and segmented handling of long audio; batch work applies the workflow to multiple inputs. Model loading supports configurable precision and quantization. The README documents these capabilities, while the partial app.py excerpts confirm key processing and save behavior; some wiring is therefore shown only where documented.
Main entry points: app.py Batch processing
Components 19
User interface
- Gradio interface
app.py - File and folder selection
app.py - Run and preprocessing settings
app.py - Model configuration
app.py
Audio preparation
- Batch processing
app.py - Audio loading
app.py - Stem separation
app.py - Noise reduction
app.py - Duration trimming
app.py - Audio segmentation
app.py
Model recognition
- Model loading
app.py - Qwen Omni recognition
app.py - Whisper transcription
app.py - Model weights
Results and saving
- Text results
app.py - Manual and auto-save
app.py - Saved result files
Other components
- User
- Audio inputs
Connections 24
- User → Gradio interface: uses
- Gradio interface → File and folder selection: offers selection
- User → File and folder selection: selects audio
- File and folder selection → Audio inputs: selects files
- Gradio interface → Run and preprocessing settings: offers settings
- Gradio interface → Model configuration: offers settings
- Run and preprocessing settings → Batch processing: starts processing
- File and folder selection → Batch processing: provides inputs
- Batch processing → Audio loading: loads audio
- Batch processing → Stem separation: optionally separates
- Batch processing → Noise reduction: optionally reduces noise
- Batch processing → Duration trimming: optionally trims
- Batch processing → Audio segmentation: optionally segments
- Model configuration → Model loading: configures loading
- Model loading → Model weights: loads weights
- Batch processing → Qwen Omni recognition: dispatches Qwen work
- Batch processing → Whisper transcription: dispatches Whisper work
- Qwen Omni recognition → Text results: returns captions
- Whisper transcription → Text results: returns transcription
- Text results → Gradio interface: displays results
- Gradio interface → Manual and auto-save: requests saving
- Batch processing → Manual and auto-save: auto-saves results
- Manual and auto-save → Saved result files: writes results
- User → Manual and auto-save: requests manual save