Detection of Audio Emotional Intelligence Using Machine Learning Algorithms

The interaction between humans and machines are increasing day by day and due to advancement in technologies the medium of audio as interaction has been growing exponentially, but unlike text medium, audio interaction needs to be more human-like to make it more effective. To reduce the gap between human and chat-bot interaction. In this research, various features such as Mel-frequency cepstral coefficients, root mean square energy, tonnetz and zero crossing rate are extracted and analysed to show which features contribute more to the identification of emotions. In addition, several machine learning models are developed and results are presented. The result of this project will help customers interact with a chatbot effectively. The ensemble model used in this project resulted in a accuracy of 67% with MFCC features which is the highest when compared to other models. After successful identification of emotions, a chatbot framework is presented which can adapt to interactive dialogues with the customer based on the emotion from the speech in the audio. In addition to this, the results of the previous related work and research gaps are discussed. To fill the research gaps feature manipulations and models are developed