Bangali Accent Recognition from Speech

Abstract

Communication between human and computer is the most vital term of current world and the performance of a computer or machine mostly depends on this communication. Undoubtedly, human speech is the most comfortable form of communication for humans and that’s why the world is now trying to reduce the dependencies on a text by using speech communication which producing a huge amount of audio data. Human speech is nothing, but analog signals, the different wavelength of this signal represents different age’s speaker, different gender’s speaker even different language’s word has a different wavelength. And the question was “dose different accent of the same language produces different wavelengths”? we tried to find out the answer to this specific question. For that, we used different accents of the Bengali language. The services of voice-based applications are very limited in the Bengali language for maximizing the benefit of voice-based applications in Bengali we need to train our machine with local Bengali language and for that we need to perfectly identify that which accent actually speaker is speaking. But the problem is these digital machines don’t handle the analog signals. That’s why we need to convert the signal into numeric value for that we used very popular and effective techniques of feature extraction which is Mel Frequency Cepstral Coefficient and for classification we used different classification algorithms. We got maximum 86% accuracy on 9303 data of different classes.