Title: Speaker Verification Based on Pitch Classified Feature Mapping
Abstract: In order to solve the problem of how to train the SVM speaker model properly while giving large quantities of speech features in text-independent speaker verification,an approach based on pitch classified GMM-UBM feature mapping and SVM is proposed,which classifies speech in the feature space by the pitch parameter and then carries out feature mapping by using GMM-UBM structure,thus to get the speaker feature and build SVM speaker models in each feature sub-space.The pitch classified feature mapping not only condenses the quantity of speech cepstral feature,but also optimizes the discriminability of SVM.Therefore,after linearly fusing the scores of sub-systems,a better speaker verification performance is obtained.The experiments in the database of NIST'06 indicate that this approach is fairly efficient.
Publication Year: 2010
Publication Date: 2010-01-01
Language: en
Type: article
Access and Citation
AI Researcher Chatbot
Get quick answers to your questions about the article from our AI researcher chatbot