Speech Emotion Recognition Using CNN and LSTM

In recent years, the demand for speech emotion recognition has intensely increased. Emotions are integral part of speech which is essential for interactive communication. However, it has been challenging task for researchers to detect exact emotion from speech for Human Computer Interaction (HCI). The Ability of Machine to detect emotions from speech will largely benefit the domain of HCI. Machine learning algorithms are often used to solve a variety of recognition problems including image, speech, and music recognition. This paper shows implementation of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) algorithms for speech emotion recognition application on EMO-DB dataset for neutral, angry, sad, and happy emotions. The average accuracy observed for CNN is 78.75% and for LSTM is 85.5%.