Requirements background
Professional music production often requires separate vocal and backing tracks, and SongGen's two-track mode is designed for this scenario.
Specific steps
- optionSongGenDualTrackForConditionalGenerationModel classes are initialized
- When generated, it gets a separatevocal_sequencescap (a poem)acc_sequencesaudio array
- Use soundfile to save them separately:
sf.write('vocal.wav', vocal_array, sr)
sf.write('accompaniment.wav', acc_array, sr)
advanced application
- Noise reduction/modulation of vocals possible
- Loudness equalization or section editing of accompaniment
- Final blend export using DAW software
This answer comes from the articleSongGen: A Single-Stage Autoregressive Transformer for Automatic Song GenerationThe
May not be reproduced without permission:AI productivity tools " How do I realize the need for independent editing of vocals and backing tracks?
