Emotion recognition of audio/speech data using deep learning approaches
Vedika Gupta, Stuti Juyal, Gurvinder Singh, Chirag Killa, Nishant Gupta
Abstract
Speech has been the most prominent and intelligent mode of sound. It is an effective medium to communicate the emotions and attitude in a particular language. Researchers have been extensively using speech to understand the emotions of a person. Contemplating emotions by intonation and pitch of voice and relative loudness can be used for enhancing human computer interaction. In this work, we have surveyed feature extraction techniques based on prosodic features or spectral feature extraction such as MFCC, LPCC, LPC, etc. and found out that MFCC extracts the best features for recognition of emotional content. This research work utilizes some of the best existing classification techniques for recognizing human emotions and conducts a detailed comparative analysis based on statistical and mathematical results. Finally, this paper proposes an optimal model DSCNN for raw spectrogram that resulted in an un-weighted accuracy of 61% for raw spectrogram (with noise) and 79% for clean spectrogram (without noise) for the enhancement of the human emotion evaluation system.