Distance Based Single-Channel Target Speech Extraction

TL;DR

The paper addresses single-channel target speech extraction in enclosed spaces by exclusively leveraging distance information, without using speaker physiological cues. It introduces a distance-based model that fuses distance cues with time-frequency bin representations to guide extraction. The approach is positioned as the first to rely solely on distance cues for single-channel TSE and is validated in both single-room and multi-room configurations, with an additional capability to estimate speaker distances from mixed speech. These results demonstrate a feasible distance-driven approach with practical implications for room-aware speech processing, and an online demo is provided.

Abstract

This paper aims to achieve single-channel target speech extraction (TSE) in enclosures by solely utilizing distance information. This is the first work that utilizes only distance cues without using speaker physiological information for single-channel TSE. Inspired by recent single-channel Distance-based separation and extraction methods, we introduce a novel model that efficiently fuses distance information with time-frequency (TF) bins for TSE. Experimental results in both single-room and multi-room scenarios demonstrate the feasibility and effectiveness of our approach. This method can also be employed to estimate the distances of different speakers in mixed speech. Online demos are available at https://runwushi.github.io/distance-demo-page.