Improving the Mean Shift Clustering Algorithm for Universal Background Model (UBM).

Speech clustering is an unlabeled technique that can find the previous information without any clustering results regarding the number of previous speakers. When the original speech information is transformed into the form of mel frequency cepstral coefficients, the transformation methodology is better standardized to represent the speaker. It is stated that each speech area is trained by a giant Gaussian mixture model (GMM) that is common to the universal background model (UBM). The UBM is a very large GMM that is trained and developed by different channels across the data. The current mean shift system works for different types of speakers from the given speech information. In the proposed work, the multi-viewpoint references are carried out by a single view instead of a single point. This type of comparison quantity is called multi-viewpoint-based comparison, currently it is known as multi-viewpoint-based similarity (MVS). Then the proposed system is developed by the optimal clustering result MVS. This technique of MS is the seeking procedure for statistical review mode. MVS functionality is higher than the other systems MS. The proposed system is capable of detecting speech clustering trend and respective speech clustering results. In this paper, the clustering validity index measurement is validated by different clustering results and shows the correctness of the results. [ABSTRACT FROM AUTHOR]