Apollo is a novel music restoration method designed to address distortions and artefacts caused by audio codecs, especially at low bitrates. Operating in the frequency domain, Apollo uses a frequency band-split module, band-sequence modeling, and frequency band reconstruction to restore the audio quality of MP3-compressed music. It divides the spectrogram into sub-bands, extracts gain-shape representations, and models both sub-band and temporal information for high-quality audio recovery. Trained with a Generative Adversarial Network (GAN), Apollo outperforms existing SR-GAN models on the MUSDB18-HQ and MoisesDB datasets, excelling in complex multi-instrument and vocal scenarios, while maintaining efficiency.
The open-sourced content includes models for inference at
https://github.com/ZFTurbo/Music-Source-Separation-Training and the original weights with fewer training steps. The training was conducted using sucial's project at
https://github.com/SUC-DriverOld/Apollo-Training, with a 92-hour high-quality vocal dataset trained for 1 million steps.
