I-vector Based Emotion Recognition in Assamese Speech

As emotion is an integral part of speech and is strongly related to a speaker’s characteristics, it plays a vital role in any speech or speaker recognition system. Assamese is a widely spoken language in the north-eastern part of India and is known for its dialectal richness and ethnographic diversity. A speech or speaker recognition system is expected to deal with the emotion content of the samples. Here, we report the design of an emotion recognition system in Assamese language exploiting the ability of the Recurrent Neural Network (RNN) to track the temporal variations in the speech sample. RNN is an Artificial Neural Network (ANN) with feed forward and feedback sections enabling it to capture time dependentvariations in speech samples. We have designed i-vector and Mel Frequency Cepstral Coefficients delta (MFCC-delta) features and report the comparative performance, for the recognition of various moods, derived from the RNN and Distributed Time Delay Neural Network (DTDNN) classifiers.