Integration of Latent Prosody Analysis and Latent Acoustics Analysis for Robust Speaker Verification

• Main Authors: Zhi-Ren Zeng, 曾志仁
• Other Authors: 廖元甫
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/spx574

碩士 === 國立臺北科技大學 === 電腦與通訊研究所 === 94 === Channel and handset mismatch is the major source of performance degradation for speaker verification in telecommunication environment. This thesis discusses the problem of compensating the channel and handset mismatch with few available train or test data, a latent prosody analyses (LPA) and a latent acoustics analysis (LAA) approaches were proposed and fused together for robust speaker verification. The proposed methods are evaluated on the standard one speaker detection task of the 2001 NIST Speaker Recognition Evaluation Corpus where only one 2-minute training and 30-second trial speech (in average) are available. Experimental results have shown that the proposed LPA+LAA fusion approach could improve the equal error rates (EERs) of MAP-GMMs+MVA+T-norm approaches from 8.9% to 6.6%. Therefore, the proposed approaches are promising and worthy further studying for real-life speaker verification systems.