Towards an Optimal Speaker Modeling in Speaker Verification Systems using Personalized Background Models

Abstract

This paper presents a novel speaker modeling approachfor speaker recognition systems. The basic idea of this approach consists of deriving the target speaker model from a personalized background model, composed only of the UBM Gaussian components which are really present in the speech of the target speaker. The motivation behind the derivation of speakers' models from personalized background models is to exploit the observeddifference insome acoustic-classes between speakers, in order to improve the performance of speaker recognition systems. The proposed approach was evaluatedfor speaker verification task using various amounts of training and testing speech data. The experimental results showed that the proposed approach is efficientin termsof both verification performance and computational cost during the testing phase of the system, compared to the traditional UBM based speaker recognition systems. Keyword:Speaker modeling Speaker verification Gaussian mixture models Universal background model Maximum a posteriori (MAP) Personalized background models

Towards an objective comparison of feature extraction techniques for automatic speaker recognition systems

“…The feature extraction component involves the processing of speech signal and the extraction of speaker-specific and discriminative characteristics as shown in Figure 1. The modeling & scoring block aims to train a reference model for each client speaker on the basis of its extracted features, as well as, to score the test utterances [1,2].…”

An optimized method towards formal verification of mixed signals using differential fed neural network over FFNN

“…From the outcome analysis [12] found an effective reduction in simulation time of SoC with respect to other existing works. Similarly for application prospective Bouziane et al [13] presented an optimal verification for speaker modeling by using personalized background models. Similar kind of work is found in Kumari and Jayanna [14] which is meant for verification of limited data speakers.…”