As, is the feature vector speech segment usedto made model . ?Ac?is the cardinality of a set and ?W?is the weighted sum of covariance matrices Cov (Ac). So, only thepartition is selected during the final clustering this result in minimum WCDscore 18.104.22.168 Issues Relatedwith AlgorithmThere are many otherissues need attention in speaker clustering.
A uniform model for all segments fromthe cluster belonging could be built called average linkage but this is quiteexpensive in terms of computational costs and furthermore suggest that someother form of linkage for segments or model in the cluster may even moresuitable-complete linkage used to compute the distance between the two clusterfor individual points.(Sander and Ester, 2000)single linkage on other hand use the distance of the nearest pair of each torepresent the distance of pair.GMMtraining process should be initializing with some reference point . The expected maximization (EM)algorithm will help to identify a local maximum likelihood regardless of thestarting point but likelihood equation for GMM has many starting point andmaxima models that give different maxima K-mean and K-mean++ are some of the initializationemployed but unfortunately maximum of them are not up to the mark and take lotsof iterations to coverage 74.
Asfrom above Fig. 2.5 when training anodal variance GMM it has been found that variance elements become small inmagnitude which is particularly true for a mixture model with a large number ofcomponent densities (?32). Such small variances generates a singularity insidethe likelihood function of model and Detroit identification performance bydistorting speaker model score used in maximum likelihood classifier. To avoidthis problem which will cause numerical instability, a maximum variance valueon elements of all variance vectors is added in a speaker’s model.
2.6 Training ofAcoustic ModelAcousticmodel is used to collect speech feature numerical data in large quantities expressedin term of parameters and is very important in refinement of various speechclasses and acoustic model in speech recognition build on the base of HiddenMarkov Models (HMM). Acoustic model are used in evaluating probability fromspeech to an acoustic unit or an acoustic hypothesis and language model areused to identify the probability of word sequence. Most of the recognitionsystems follow HMM as the acoustic modelling rule.