Learning aided mood and dialect recognition using telephonic speech

Speech is the most basic means of human communication and contains information regarding message, speaker's identity, dialects, mood and so on. Emotion is an integral part of speech that plays a vital role in the improvement of speech recognition system. The task of dialect recognition also helps in improving the automatic speech recognition (ASR) system for more natural communication between human and machines. In our paper, we report the design of a mood and dialect recognition system for Assamese telephonic speech using Fourier parameters. The Mel Frequency Cepstral Coefficients (MFCC) and its first order derivatives and also the Fourier parameters are derived and the comparative performance for the recognition of various moods and major dialects of Assamese language obtained from Recurrent Neural Network (RNN), Distributed Time Delay Neural Network (DTDNN) and Support Vector Machine (SVM) is reported here.