Phoneme Classification Using Temporal Tracking of Speech Clusters in Spectro-temporal Domain

Abstract

This article presents a new feature extraction technique based on the temporal tracking of clusters in spectro-temporal features space. In the proposed method, auditory cortical outputs were clustered. The attributes of speech clusters were extracted as secondary features. However, the shape and position of speech clusters change during the time. The clusters temporally tracked and temporal tracking parameters were considered in secondary features. The new architecture was proposed for phoneme classification by a combining classifier using both tracked and energy-based features. Clustered based spectro-temporal features vectors were used for the classification of several subsets of TIMIT database phonemes. The results show that the phoneme classification rate was improved Using tracked spectrotemporal features. The results were improved to 78.9% on voiced plosives classification which was relatively 3.3% higher than the results of non-tracked spectro-temporal feature vectors. The results on other subsets of phonemes showed good improvement in classification rate too.

A Clustering-Based Approach for Features Extraction in Spectro-Temporal Domain Using Artificial Neural Network

“…Therefore, in recent researches, clustering methods were used to reduce dimension of spectro-temporal features space and extract valuable discriminative information of speech signal. In these methods, output of this model was considered as the primary features vectors and clustered using Gaussian Mixture Model and weighted K-Means [14][15][16][17]. Then, *Corresponding Author Institutional Email: na_esfandian@Qaemiau.ac.ir (N. Esfandian) the mean vectors and covariance matrices elements of the clusters are considered as secondary features in each speech frame.…”