Neural Networks: A Comprehensive Foundation
Neural Networks: A Comprehensive Foundation
A Two Stage Procedure for Phone Based Speaker Verfication
AVBPA '97 Proceedings of the First International Conference on Audio- and Video-Based Biometric Person Authentication
Improving a GMM speaker verification system by phonetic weighting
ICASSP '99 Proceedings of the Acoustics, Speech, and Signal Processing, 1999. on 1999 IEEE International Conference - Volume 01
| Hi-index | 0.00 |
Traditional speaker verification systems are limited to the use of frame-based spectral features that are basically modeled globally via Gaussian Mixture Models (GMM). With such methods the probability density function of the acoustic feature vectors is estimated globally and the linguistic structure of the speech signal is not taken into account. In this paper we study the performance of a speaker verification system based on a combination of a data-driven Automatic Language Independent Speech Processing (ALISP) segmentation and a classical GMM based system. Even though the ALISP classes are not being explicitly modeled by the GMMs and the segmental information is used only during the scoring phase, the proposed segmental approach slightly outperforms the baseline global GMM system. Two techniques are used to combine the segmental scores in order to exploit the different amounts of discrimination provided by the ALISP classes: the Logistic Regression and the Multi-Layer Perceptron. Improvement in performance has been made by using the Multi-Layer Perceptron. The evaluation of the proposed method is done on the NIST 2004 Speaker Recognition Evaluation data.