Extreme learning machine based speaker recognition
Zongjiang. Hu · DR-NTU (Nanyang Technological University) · 2011
Speaker Recognition is the process of automatically recognizing who is speaking based on the individual information included in speech waves. Speaker Recognition is basically classified into speaker identification and speaker verification. It can be used in many fields such as banking by telephone, voice dialling, voice mail, and security control for secret information areas. Extreme Learning Machine (ELM), originally proposed by Prof. Huang, is an emerging technique which produces very good generalization performance in regression applications as well as in large dataset classification applications. In this report, a text-independent close-set speaker verification system based on ELM and SVM will be constructed and each part of the system like speaker database, feature extraction, speaker modelling and decision making will be elaborated in detail. Finally and most importantly, the performance of ELM and SVM will be compared. With the results shown in Chapter 6, we can draw the conclusion that ELM is superior to SVM in terms of tuning simplicity, time efficiency and testing accuracy.