TISDiSS: A Training-Time and Inference-Time Scalable Framework for Discriminative Source Separation
Abstract
Source separation is a fundamental task in speech, music, and audio processing, and it also provides cleaner and larger data for training generative models. However, improving separation performance in practice often depends on increasingly large networks, inflating training and deployment costs. Motivated by recent advances in inference-time scaling for generative modeling, we propose Training-Time and Inference-Time Scalable Discriminative Source Separation (TISDiSS), a unified framework that integrates early-split multi-loss supervision, shared-parameter design, and dynamic inference repetitions. TISDiSS enables flexible speed-performance trade-offs by adjusting inference depth without retraining additional models. We further provide systematic analyses of architectural and training choices and show that training with more inference repetitions improves shallow-inference performance, benefiting low-latency applications. Experiments on standard speech separation benchmarks demonstrate state-of-the-art performance with a reduced parameter count, establishing TISDiSS as a scalable and practical framework for adaptive source separation. Code is available at https://github.com/WingSingFung/TISDiSS.
Cite
@article{arxiv.2509.15666,
title = {TISDiSS: A Training-Time and Inference-Time Scalable Framework for Discriminative Source Separation},
author = {Yongsheng Feng and Yuetonghui Xu and Jiehui Luo and Hongjia Liu and Xiaobing Li and Feng Yu and Wei Li},
journal= {arXiv preprint arXiv:2509.15666},
year = {2025}
}
Comments
Submitted to ICASSP 2026.(C) 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work