Single-Channel Target Speech Extraction Utilizing Distance and Room Clues

Author: Runwu Shi, Zirui Lin, Benjamin Yen, Kazuhiro Nakadai

Abstract: This paper aims to achieve single-channel target speech extraction (TSE) in enclosures utilizing distance cues and room information. Recent research has verified the feasibility of distance cues, which implies the sound source’s direct-to- reverberation ratio (DRR) and thus can be utilized for speech separation and TSE systems. However, such distance cue is significantly influenced by the room acoustic environment such as dimension and reverberant time, making it challenging for TSE systems that rely solely on distance cues to generalize across a variety of different rooms. To solve this, we sug- gest providing room environmental information for distance- based TSE for better generalization capabilities. Especially, we propose a distance and environment-based TSE model in the time-frequency (TF) domain with learnable distance and room embedding, and the results on both simulated and real collected dataset demonstrate its feasibility.