Music Source Separation with Generative Flow
arXiv:2204.09079 · doi:10.1109/LSP.2022.3219355
Abstract
Fully-supervised models for source separation are trained on parallel mixture-source data and are currently state-of-the-art. However, such parallel data is often difficult to obtain, and it is cumbersome to adapt trained models to mixtures with new sources. Source-only supervised models, in contrast, only require individual source data for training. In this paper, we first leverage flow-based generators to train individual music source priors and then use these models, along with likelihood-based objectives, to separate music mixtures. We show that in singing voice separation and music separation tasks, our proposed method is competitive with a fully-supervised approach. We also demonstrate that we can flexibly add new types of sources, whereas fully-supervised approaches would require retraining of the entire model.
arXiv commentAccepted by Signal Processing Letters
References in corpus (15)
- Adam: A Method for Stochastic Optimization
- Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation
- Supervised and Unsupervised Speech Enhancement Using Nonnegative Matrix Factorization
- Glow-TTS: A Generative Flow for Text-to-Speech via Monotonic Alignment Search
- Music Source Separation in the Waveform Domain
- Wave-U-Net: A Multi-Scale Neural Network for End-to-End Audio Source Separation
- Hybrid Spectrogram and Waveform Source Separation
- Music Demixing Challenge 2021
- The 2018 Signal Separation Evaluation Campaign
- Decoupling Magnitude and Phase Estimation with Deep ResUNet for Music Source Separation
- Source Separation with Deep Generative Priors
- CatNet: music source separation system with mix-audio augmentation
- Single-Channel Signal Separation and Deconvolution with Generative Adversarial Networks
- Transfer Learning with Jukebox for Music Source Separation
- Problems using deep generative models for probabilistic audio source separation