Abstract: As the most natural interface for human interaction, speech can be exploited to track users and then customize services as they get available. Low latency is required, since adaptation to user profiles must be done in a continuous fashion. However, most speaker tracking approaches found in the literature work offline, fully processing pre-recorded audio files by means of a two-stage procedure involving acoustic segmentation and speaker detection. In this work, a low-latency online speaker tracking approach is applied, which deals with continuous audio streams and outputs a decision at fixed intervals, by scoring fixed-length audio segments with regard to a set of target speaker models. Experimental results are reported on the AMI Corpus of meeting conversations, revealing the effectiveness of the proposed approach with regard to a traditional approach working offline. A speaker tracking service and a lower-level auxiliary speaker detection service have been also designed, based on the online low-latency speaker tracking approach mentioned above. These services are SOA-compliant and provide an interoperable, reusable and easily evolvable means to develop SOA-based speaker tracking applications for Ambient Intelligence (AmI) environments.
Index Terms: low-latency, speaker tracking, Service Oriented Architecture, Ambient Intelligence.