| Summary: | 碩士 === 義守大學 === 電機工程學系 === 90 === In recent years, with the rapid development of telephone voice system, wireless communication and e-commerce system, the security problem of information and internet is getting more and more serious. Facing this situation, speaker recognition is probably the most appropriate solution and the related technology is nowadays rather mature. Generally speaking, the accuracy of speaker recognition is satisfactory, but there still exist some problems concerning system integration. For example, it is not easy to implement with low cost because of the complexity of system algorithm, so the goal of being popularized is hard to reach. Therefore, we propose in this paper a Hybrid Neural Network (SOM-PNN) classifier with high performance and simple structure, substitute for the Gaussian Mixture Model (GMM) with high complexity and computation. This classifier is composed of a Self-Organizing Map (SOM) neural network and a Probabilistic Neural Network (PNN). According to our experiment result, the Hybrid Neural Network classifier presents better performance, higher calculating speed and less memory requirements of that than the GMM classifier. As this classifier’s algorithm is simple and its requirement for memory is substantially reduced, the hardware implementation will be easily carried out.
|
|---|
