Emotion recognition on the basis of audio signal using Naive Bayes classifier

Abstract

In this paper we have studied and implemented the classification of audio signal into four basic emotional state. For that we have considered different statistical features of pitch, energy, and ZCR (Zero Crossing Rate) MFCC (Mel frequency cepstral coefficient) from 2000 utterances of the created audio signal database. In that, Pitch feature is extracted by AMDF (average magnitude difference method) and energy is calculated by sum of square absolute value of magnitude spectrum. And MFCC is calculated by taking DCT (Discrete cosine transform) of its energies spectrum by keeping the DCT coefficients 1-14 and discarding the rest. In statistical modeling, regression analysis is a statistical process for calculating approximately the variables. It comprise many techniques for modeling and analyzing several variables. In this paper Naïve Bayes Classifier is used to classify the audio signal into four different emotions. Speech signal is random signal so we have to predict the future sample and Naïve Bayes Classifier is totally probability based classifier so in speech analysis for accurate prediction we are using Naïve Bayes classifier. In the speech signal for recognition of signal classifier require millions of dataset. The advantage of Naïve Bayes classifier is that it recognizes the signal with minimum dataset.