An Online Speaker Tracking System for Ambient Intelligence Environments

Abstract

Ambient intelligence is an interdisciplinary paradigm which envisages smart spaces that provide services and adapt transparently to the user. As the most natural interface for human interaction, speech can be exploited for adaptation purposes in such scenarios. Low latency is required, since adaptation must be continuous. Most speaker tracking approaches found in the literature work offline, fully processing prerecorded audio files by a two-stage procedure: (1) performing acoustic segmentation and (2) assigning each segment a speaker label. In this work a real-time low-latency speaker tracking system is presented, which deals with continuous audio streams. Experimental results are reported on the AMI Corpus of meeting conversations, revealing the effectiveness of the proposed approach when compared to an offline speaker tracking system developed for reference.