Feature Comparison for Speech Emotion Recognition on Hindi Language

Emotions have the power to change the meaning and context of delivered speech. The current study provides an overview of the impact of distinct speech features on vocal based emotion recognition model. The effect of the a forementioned attributes in speech has been tested and validated using a variety of local features. Various emotions—Angry, Neutral, Happy, Sad, Surprise, Sarcasm and Disgust—have been represented by male and female voices, and their performances have been assessed. In addition, a well-known deep learning model named LSTM, have been used on the Hindi dataset IITKGP: SEHSC. The performance comparison of various set of speech features is performed on various measures such as Accuracy, Precision and F-1 Score. However, the best accuracy of 80% on IITKGP: SEHSC datasets is achieved when considering all the features together as an input set.