Speech parameterization based on AM-FM model and its application in speaker identification using AANN

This paper presents the parameterization of speechbased on amplitude and frequency modulation(AM-FM) model and its application to speakeridentification. Speech parameterization is based onthree different bandwidths viz 400Hz, 266mel,106mel. The feature obtained by thisparameterization is termed as PYKFEC which is notdirectly used as a feature instead its average of eachfilter is used as the feature and termed as FAP. Thespeaker identification is done using auto associativeneural network and Gaussian mixture model. TheAANN/GMM is trained using the SOLO speakingstyle from CHAINS CORPUS database and anetwork/model is created for each speaker. Thecreated model is tested using different speaking stylelike FAST and WHSP of the speaker. Theidentification rate of FAP is better than PYKFEC,and AANN performs well with these features.