Comparative evaluation of maximum a Posteriori vector quantization and gaussian mixture models in speaker verification

• Published in: Pattern recognition letters Vol. 30; no. 4; pp. 341 - 347
• Main Authors: Kinnunen, Tomi, Saastamoinen, Juhani, Hautamäki, Ville, Vinni, Mikko, Fränti, Pasi
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-03-2009; Elsevier
• Subjects: Applied sciences; Coding, codes; Exact sciences and technology; Gaussian mixture model (GMM); Information, signal and communications theory; MFCC features; Sampling, quantization; Signal and communications theory; Signal processing; Signal representation. Spectral analysis; Signal, noise; Speaker verification; Speech processing; Support vector machine (SVM); Telecommunications and information theory; Universal background model (UBM); Vector quantization (VQ); Universal background model (UBM); Support vector machine (SVM); Gaussian mixture model (GMM); Vector quantization (VQ); Speaker verification; MFCC features; Performance evaluation; Discriminant analysis; Automatic classification; Vector quantization; Background; Data compression; Mixture theory; Support vector machine; Speaker recognition; Signal classification; Learning; Credal approach; Cepstral analysis; Gaussian process; A posteriori estimation; Maximum likelihood; Speech processing
• ISSN: 0167-8655; 1872-7344
• DOI: 10.1016/j.patrec.2008.11.007

Gaussian mixture model with universal background model (GMM–UBM) is a standard reference classifier in speaker verification. We have recently proposed a simplified model using vector quantization (VQ–UBM). In this study, we extensively compare these two classifiers on NIST 2005, 2006 and 2008 SRE corpora, while having a standard discriminative classifier (GLDS–SVM) as a point of reference. We focus on parameter setting for
N-top scoring, model order, and performance for different amounts of training data. The most interesting result, against a general belief, is that GMM–UBM yields better results for short segments whereas VQ–UBM is good for long utterances. The results also suggest that maximum likelihood training of the UBM is sub-optimal, and hence, alternative ways to train the UBM should be considered.