A Class of Neuro-computational Models to Verify Mood Variation in Dialectal Assamese Speech

Mood content in spoken word recognition is an important element in formulation of a decision support system (DSS). Many times it becomes integral components of human computer interaction (HCI) systems based on speech recognition with language orientation. In this paper, we propose a mood verification system of speakers of Assamese language with dialectal components. Five features namely Mel Frequency Cepstral Coefficients (MFCC), Linear Predictive coding (LPC), Singular Value Decomposition (SVD), Principal Component Analysis (PCA) and a composite set comprising of all the four features mentioned above have been used with Recurrent Neural Network (RNN) and Feed forward Time Delay Neural Network (FFTDNN) to evaluate their performance in recognizing mood variations in dialectal Assamese. The system has been tested under several different background noise conditions by considering the recognition rates and computation time.