Mario Refice

The Influence of Frame Length on Speaker Identification Performance

Speaker recognition/identification is a challenge for the implementation of security applications. Unfortunately, degradation in performance is usually observed for high pitched speakers and also whenever average pitch varies significantly between enrolment and testing. In this paper, a study on the impact of the frame length used to extract features from speech signal on the performance of speaker identification is presented. Tests have been carried out on a text-dependent database. Results show that a combination of different frame sizes between the training and the recognition phases can cope with the degradation. A reduction between 40% and 65% in false rejections has been generally observed.

Handwritten Signature and Speech: Preliminary Experiments on Multiple Source and Classifiers for Personal Identity Verification

This paper presents a personal verification system based on two different biometric traits: handwritten signature and speech. The signature verification system uses contour-based features and a Dynamic Time Warping technique for matching. The speaker verification system uses cepstral based coefficients and is based on a Hidden Markov Model statistical classifier. In the decision combination stage, the decisions provided by the two systems are combined according to a simple abstract---level combination approach. The experimental results related to a real-scenario demonstrate the effectiveness of the proposed approach and highlight some profitable directions for further developments.

Speaker Identification by Multi-Frame Generative Models

In this paper an approach called multi-frame speaker models (MFS) is proposed, in order to cope with performance degradation generally observed over (short and medium) time and trials in speaker identifications task. The approach, based on generative models, uses multiple frames length for speech processing in training and testing phase. A complete multi-expert system is also presented which is able to implement the proposed approach onthe whole set of speakers and to obtain a near optimum for the ERs reduction.

Pseudo Multi Parallel Branch HMM for Speaker Verification

This paper describes an approach, called Pseudo Multi Parallel Branch (P-MPB) Model, for coping with performance degradation typically observed over (short and medium) time and trials in Text Dependent Speaker Verification’s task. It is based on the use of a fused HMM model having a multi parallel branch topology where each branch consists of an HMM referring to a specific frame’s length representation of the speaker. The approach shows reasonable ER’s reduction from the baseline system.

Improving Text-Dependent Speaker Recognition Performance

In this paper we investigated the role of the frame length on the computation of MFCC acoustic parameters in a text-dependent speaker recognition system. Since the vocal characteristics of subjects may vary along the time, the related information conveyed by the MFCCs usually cause a significant degradation on recognition performance. In our ex- periment we tested the use of different frame lengths for the features extraction in the training and the recognition phases for a set of speakers whose speech productions spanned over 3 months. Results show that a suitable choice of the frame lengths combination for training and testing phases can improve the recognition performance reducing the false rejection rate. An expert system driven to look for the best combination of frame lengths in order to obtain the maximum performance level of the HHM engine may help in decreasing the amount of false rejections.