K. Chen, L. Wang, and H. Chi. Methods of combining multiple classifiers with different features and their applications to text-independent speaker identification. Int. J. of Pattern Recognition and Artificial Intelligence, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and when the score is below but close to the threshold, or in case that more than one speaker models produce scores above the threshold, the GMM classifier is called to contribute to the final decision. The scores of the two classifiers are weighted using the linear combination method described in [8]. In both the cases of multi class and binary classifiers, the final decision is made after applying the speakerindependent final threshold. Because the procedures for extracting the final decision in the Speaker Identification and the Speaker Verification systems differ significantly, they are ....

Ke Chen, Lan Wang and Huisheng Chi, "Methods of Combining Multiple Classifiers with Different Feature and Their Applications to Text-Independent Speaker Identification", International Journal of Pattern Recognition and Artificial Intelligence, 11(3), pp.417-445, 1997.

....the best classification system is selected on the basis of an evaluation. Recently, the possibilities of combining sets of classifiers has been considered. There are many examples found in which such a combination of classifiers has a better performance than any of the base classifiers in the set [3,9,11]. How to construct such a combination of classifiers has become an important direction of research [14,15] The difficulties that arise in combining a set of classifiers is directly evident if one considers the metaphor of a committee of experts. How has such a committee to arrive at a final ....

K. Chen, L. Wang and H.S. Chi, Methods of combining multiple classifiers with different features and their applications to text-independent speaker identification, International Journal of Pattern Recognition and Artificial Intelligence, vol. 11, no. 3, 417-445.

....for speaker verification. The suprasegmental features are robust against transmission channel variations. For real applications, it may be useful to combine evidences from different approaches. It is believed that by combining multiple classifiers, the classification performance can be improved [2, 3]. This paper consists of two parts. The first part describes a neural networkbased text dependent speaker verification system using suprasegmental features. In the second part of this paper, we have proposed an approach to combine the evidence present at the segmental and suprasegmental levels ....

....text dependent speaker verification system using suprasegmental features. In the second part of this paper, we have proposed an approach to combine the evidence present at the segmental and suprasegmental levels within the framework of evidential theory for reliable speaker verification [2, 3]. 2 Development of the Speaker Verification System Speaker verification task consists of three stages: Feature extraction, feature comparison and decision making. It has two operational phases: Training and testing. During training, a speaker specific model template is generated from the ....

K. Chen and L. Wang and H. Chi, "Methods of combining multiple classifiers with different features and their application to text-independent speaker identification," International Journal of Pattern Recognition and Artificial Intelligence, Vol. 11, no. 3, pp. 417--445, 1997.

....community and Z. yields improved performance Xu et al. 1992; Suen et al. 1993; Ho et al. 1994; Huang and Suen, 1995 . Furthermore, more recent studies show that better performance can be achieved by combining multiple Z. classifiers with diverse features Xu et al. 1992; Perrone, 1993; Chen et al. 1997 in contrast to the combination of multiple classifiers with the same feature. For most of the combination methods, however, a large amount of data is usually required to train both individual classifiers and a combination scheme. In addition, the combination methods are also viewed as a kind of ....

....been reported that the BKS method achieves a promising performance and outperforms several classical combination methods in an unconstrained handwritten numerals Z. recognition problem Huang and Suen, 1995 , while the BAYES method also readily yields good performance Z. Z in speaker identification Chen et al. 1997 and the hand written optical character recognition Xu et al. 1992; Perrone, 1993 . In simulations, the aforementioned HME classifiers trained on different feature sets were used as individual classifiers. The cross validation set was used to calculate the confusion matrix in the BAYES Z. Z ....

Chen, K., Wang, L., Chi, H., 1997. Method of combining multiple classifiers with different features and their applications to Z. text-independent speaker recognition. Internat. J. Pattern Recogn. Artif. Intell. 11 3 , 417--445.

No context found.

K. Chen, L. Wang, and H. Chi. Methods of combining multiple classifiers with different features and their applications to text-independent speaker identification. Int. J. of Pattern Recognition and Artificial Intelligence, 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC