Speech Emotion Recognition for Indonesian Language Using Long Short-Term Memory

Abstract

This paper presents an extended research of emotion recognition system for Indonesian language. In this research we use Indonesian Emotional Corpus with four emotions classes (anger, contentment, happiness, sadness) and neutral class. As all previous researches for emotion recognition for Indonesian language are using SVM, we are using SVM as baseline. Support Vector Machine (SVM), Feed Forward Neural Network (FFNN) and Long Short-Term Memory (LSTM) are experimented to model emotions. Experiment result shows that LSTM outperform SVM and FFNN. LSTM obtain 65.9% for average F1 measure with using acoustic and lexical feature, making it 5% higher than the best SVM in this experiment.