Effective Utterance Interpretation of Acronymic Letter Prosody through MFCC-HMM Algorithm

Abstract

The era of fast and spontaneous communication do require effective utterance interpretation which in turn requires lexical, syntactic and semantic familiarity acquaintance. This acquaintance is required at acoustic, phonetic, linguistic and application, which are the speech grades. It is proven indeed that vocal mechanism of a particular speaker is unique but, still there is a utterance difference do exists with respect to speakers mood, his/her speaking style, the context of speech and individuals intentional utterance control, perturbations and frame of his/her mind. Acronymic letter prosody uses characteristics such as pattern, accentuation, pitch and tonality to convey information and meaning pertaining to uttered letter or word. Acronymic letter prosody recognition is a challenging task in all languages. A three letter approach, heuristic approach, logistical regression approach and SVM classifier and HMM are some of the previous works carried out on this aspect. In comparison with these developed approaches, this paper brings out an effective algorithm based on MFCC-HMM approach in order to arrest variability such as to get good accuracy. The work is implemented using MATLAB platform with feature extraction using MFCC, dimensional reduction through principle component analysis (PCA), Modelling and classification through HMM.