30+ Years of Source Separation Research: Achievements and Future Challenges

TL;DR

This survey addresses the problem of separating multiple acoustic sources from mixtures byCLASSifying and evaluating model-based, deep learning, and hybrid approaches across speech, audio, and music domains. It highlights foundational methods (ICA, IVA, NMF), modern deep learning advances (PIT, deep clustering, time-domain models like Conv-TasNet), and the rise of hybrid approaches that fuse learning with beamforming. The paper also documents community-driven initiatives (SiSEC, CHiME, MDX/SDX), standard metrics (SDR, SIR, SAR, SI-SDR) and datasets (CHiME, MUSDB18, LibriMix) that have shaped reproducibility, while acknowledging gaps in real-world generalization and unknown source counts. It points to future directions including unsupervised/semi-supervised learning, diffusion-based generative models, source-activity detection, moving or distributed arrays, and efficient edge implementations to bring SS from benchmarks to real-time, everyday use.

Abstract

Source separation (SS) of acoustic signals is a research field that emerged in the mid-1990s and has flourished ever since. On the occasion of ICASSP's 50th anniversary, we review the major contributions and advancements in the past three decades in the speech, audio, and music SS research field. We will cover both single- and multi-channel SS approaches. We will also look back on key efforts to foster a culture of scientific evaluation in the research field, including challenges, performance metrics, and datasets. We will conclude by discussing current trends and future research directions.