Exploring Features for Emotion Recognition in Thai

Automated emotion recognition is an important process towards the effective adjustment of children in their environment. Unmatured children are easily influenced by violent activities from movies and television programs. Hence, violence rating systems are developed to information about the violent levels in movies and television programs to help parent make informed choices. In Thailand, a manual violence rating system is used throughout the kingdom. However, it is subjective and slow. We propose an automatic emotion recognition process to determine the rating in Thai television programs. In this work, we adopt the Hidden Markov technique to build the emotion recognition model for Thai language. The emotions that we are interested in are happiness, sadness, anger, and neutral. In our experiment, only one Thai drama is used to determine the efficiency of Mel-frequency cepstral coefficients, which is the sound feature that we selected. Keywords—Emotion Recognition, Feature Selection