Gaussian Mixture Model (GMM) based speaker verification has been widely used recently. However, little research has been performed using GMMs for actual in-vehicle speaker verification. In this paper, we propose to integrate speaker verification and localization techniques for an in-vehicle speech dialog system to locate the desired speaker. The proposed solution is able to locate both desired and undesired speakers who are talking from the same position. This problem cannot be addressed by a simply speaker localization technique only. We demonstrate that using speech data collected in real car environments, the Equal Error rate (EER) performance approaches 0 using gender dependent data, and 2.35% and 13.34% using randomly selected data under idle and city noise environments, respectively.
A Frame Based Spoken Dialog System for Home Care
Falavigna, Giuseppe Daniele;Gretter, Roberto;
2005-01-01
Abstract
Gaussian Mixture Model (GMM) based speaker verification has been widely used recently. However, little research has been performed using GMMs for actual in-vehicle speaker verification. In this paper, we propose to integrate speaker verification and localization techniques for an in-vehicle speech dialog system to locate the desired speaker. The proposed solution is able to locate both desired and undesired speakers who are talking from the same position. This problem cannot be addressed by a simply speaker localization technique only. We demonstrate that using speech data collected in real car environments, the Equal Error rate (EER) performance approaches 0 using gender dependent data, and 2.35% and 13.34% using randomly selected data under idle and city noise environments, respectively.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
