Emotion Recognition from Speech Signals

• Performed segment-level extraction of Prosodic features from raw data (IEMOCAP dataset) using MATLAB. • Trained DNNs in Theano resulting in an emotion state probability distribution for each speech segment. • Constructed utterance level features used to train an SVM for utterance level classification. Resulted in a 20% increase in relative accuracy as compared to state-of-the-art approaches. Also investigated the use of Extreme Learning Machine (DNN-ELM combination) for emotion recognition.