Saved in:
Bibliographic Details
Main Authors: You, Zhenghai, Shi, Ying, Li, Lantian, Wang, Dong
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.10921
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915854456717312
author You, Zhenghai
Shi, Ying
Li, Lantian
Wang, Dong
author_facet You, Zhenghai
Shi, Ying
Li, Lantian
Wang, Dong
contents Target speaker extraction (TSE) aims to recover a target speaker's speech from a mixture using a reference utterance as a cue. Most TSE systems adopt conditional auto-encoder architectures with one-step inference. Inspired by test-time scaling, we propose a training-free multi-step inference method that enables iterative refinement with a frozen pretrained model. At each step, new candidates are generated by interpolating the original mixture and the previous estimate, and the best candidate is selected for further refinement until convergence. Experiments show that, when ground-truth target speech is available, optimizing an intrusive metric (SI-SDRi) yields consistent gains across multiple evaluation metrics. Without ground truth, optimizing non-intrusive metrics (UTMOS or SpkSim) improves the corresponding metric but may hurt others. We therefore introduce joint metric optimization to balance these objectives, enabling controllable extraction preferences for practical deployment.
format Preprint
id arxiv_https___arxiv_org_abs_2603_10921
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Training-Free Multi-Step Inference for Target Speaker Extraction
You, Zhenghai
Shi, Ying
Li, Lantian
Wang, Dong
Sound
Target speaker extraction (TSE) aims to recover a target speaker's speech from a mixture using a reference utterance as a cue. Most TSE systems adopt conditional auto-encoder architectures with one-step inference. Inspired by test-time scaling, we propose a training-free multi-step inference method that enables iterative refinement with a frozen pretrained model. At each step, new candidates are generated by interpolating the original mixture and the previous estimate, and the best candidate is selected for further refinement until convergence. Experiments show that, when ground-truth target speech is available, optimizing an intrusive metric (SI-SDRi) yields consistent gains across multiple evaluation metrics. Without ground truth, optimizing non-intrusive metrics (UTMOS or SpkSim) improves the corresponding metric but may hurt others. We therefore introduce joint metric optimization to balance these objectives, enabling controllable extraction preferences for practical deployment.
title Training-Free Multi-Step Inference for Target Speaker Extraction
topic Sound
url https://arxiv.org/abs/2603.10921