SLT 2026 REAL-TSE Challenge: Real-world Target Speaker Extraction from Conversational Recordings
Sound
Summary
The authors present the REAL-TSE Challenge, a competition where systems try to isolate one person's speech from real recordings of conversations with multiple speakers. Unlike previous tests based on artificial speech, this challenge uses natural Mandarin and English conversations, which include background noise and overlapping speech. There are two parts: one for quick, live processing and one for analyzing the whole recording afterward. The authors also explain how they measured success and share insights from the competing systems to help future research.
Authors
Shuai Wang, Zihan Qian, Ke Zhang, Jiangyu Han, Zikai Liu, Xiaoyang Yu, Haoyu Li, Marc Delcroix, Kai Yu, Lei Xie, Ming Li, Haizhou Li
Abstract
We introduce the REAL-TSE Challenge, an IEEE SLT 2026 satellite challenge on target speaker extraction~(TSE) from real conversational recordings. Given a multi-speaker mixture and one or more enrollment utterances from a target speaker, participating systems must recover only the target speech. Unlike simulated read-speech benchmarks, REAL-TSE evaluates Mandarin and English recordings that contain natural overlap, reverberation, noise, channel mismatch, and conversational dynamics. The challenge defines two complementary tracks: an Online track for low-latency streaming extraction and an Offline track for full-context processing. Systems are evaluated with Token Error Rate (TER), Speaker Similarity (SpkSim), DNSMOS, and target-speaker activity F1. This overview paper describes the task definition, datasets, baselines, evaluation protocol, submitted systems, condition-wise findings, and lessons for future real-world TSE benchmarks.