Showing papers by "Timothy J. Hazen published in 2002"

Asr dependent techniques for speaker identification

Abstract: Traditional text independent speaker recognition systems are based on Gaussian Mixture Models (GMMs) trained globally over all speech from a given speaker. In this paper, we describe alternative methods for performing speaker identification that utilize domain dependent automatic speech recognition (ASR) to provide a phonetic segmentation of the test utterance. When evaluated on YOHO, several of these approaches were able outperform previously published results on the speaker ID task. On a more difficult conversational speech task, we were able to use a combination of classifiers to reduce identification error rates on single test utterances. Over multiple utterances, the ASR dependent approaches performed significantly better than the ASR independent methods. Using an approach we call speaker adaptive modeling for speaker identification, we were able to reduce speaker identification error rates by 39% over a baseline GMM approach when observing five test utterances from a speaker.

Rapid Speaker Adaptation With Speaker Clustering

Abstract: This paper examines an approach to speaker adaptation called speaker cluster weighting (SCW) for rapid adaptation in the Jupiter weather information system. SCW extends the ideas of previous speaker cluster techniques by allowing the speaker cluster models (learned from training data) to be adaptively weighted to match the current speaker. We explore strategies for automatic speaker clustering as well as cluster model training procedures for use with this algorithm. As part of this exploration, we develop a novel algorithm called least squares linear regression (LSLR) clustering for the clustering of speakers for whom only a small amount of data is available.