Deep learning and audio based emotion recognition

Emotion is a complex psycho-physiological changes in the mood of an individual arising from the interaction of internal and environmental factors. Emotions play an important role in people's interactions with the outside world, in their decisions and actions. Therefore, emotion recognition and is especially important for human-computer interaction, and it takes computers beyond machines that do logical processing. Visual, auditory, tactile and other biometric signals are used to detect emotions. Among these signals, speech, which is the most common means of communication between people, has an important place. This has enabled speech-based emotion recognition applications to become a field of research that has been developed with increasing interest. In this thesis, voice processing-based emotion recognition application was carried out. Three different data sets are used as SAVEE, RML and RAVDESS. After pre-processing and feature extraction are applied to existing audio files, they are classified with Long Short-Term Memory. The emotion recognition accuracy rates of the system were found 72.13% for the SAVEE data set, 70.35% for the RML data set, and 68.8% for the RAVDESS data.