Enhanced Marathi Speech Recognition Using Double Delta MFCC and DTW

Abstract

This paper describes the technique Mel-Frequency Cepstral Coefficients (MFCC), Delta MFCC, and Double Delta MFCC for Extract the Features and DTW for Pattern Matching of Automatic Speech Recognition (ASR) for Marathi. These studies present a speaker-independent Speech Recognition System for Marathi. The Dataset of Created of Speech being natural data and Speech disordered people's speech data in the Marathi language. The dataset of speech samples was created with samples of Marathi Digits and words with and without speech disorder. „PRAAT‟ was used for these recordings. Various feature extraction techniques are available, but MFCC is widely used and here, Double Delta MFCC is used to increase the recognition rate along with DTW for Pattern Matching. Speech being natural interaction medium technology can be used to develop small interfaces which can be used for various applications to help interaction between human and computer systems. This research aims to have good interaction between the Human and Computer Systems and, the patient suffering from Speech disorders peoples. That means the "Voice Technology" has been improved in the Marathi Language.

Keywords:Automatic Speech Recognition (ASR), Double Delta Mel-Frequency Cepstral Coefficients (DD MFCC), Dynamic Time Warping (DTW), Small Vocabulary Marathi Speech (SVMS)