Audio separation improves with iterative multi-input multi-output approach
Iterative Audio Separation with Mixture Consistency via MIMO Model Extension
SoundMachine Learning
Summary
Separating different sounds from a mixed recording can be tricky, especially if you want the separated sounds to sound natural and accurate. The authors propose a method that repeats the separation process several times using a model that takes in multiple sound inputs and produces multiple outputs. This iterative approach helps keep the separated sounds consistent with the original mixture. Their experiments show that adapting current top methods with this approach leads to better sound separation.
What this means in practice
- •For audio software developers: Improve algorithms in audio editing tools to separate sound sources more accurately while preserving natural audio quality.
- •For speech recognition engineers: Enhance preprocessing to isolate speech from noisy environments more reliably using iterative multi-input multi-output models.
Authors
Yukara Ikemiya, WeiHsiang Liao, Yuki Mitsufuji
Abstract
This paper proposes a general framework for stable and effective iterative audio separation with mixture consistency by extending source separation models to a multi-input multi-output (MIMO) configuration. In the field of audio separation, mixture consistency is an essential property for many applications that require accurate phase and timbral information of target sources. While iterative approaches such as diffusion models achieve perceptually superior results in speech enhancement or user-guided target source separation tasks, most existing methods focus on single-step separation with a single-input single-output (SISO) or single-input multi-output (SIMO) configuration through architectural improvements, since mixture-consistent audio separation is generally regarded as a regression problem that admits a unique solution. By extending these architectures to a MIMO configuration, we introduce iterative prediction without compromising the architectural advantages or the characteristics of mixture consistency. We conduct a comprehensive ablation study of combining the framework with discriminators and extending it to a generative model. Experimental results demonstrate significant performance improvements when applying the proposed framework to state-of-the-art separation models.