Speech emotion analysis using convolutional neural network (CNN) and gamma classifier-based error correcting output...

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Speech Processing

Background:

• Speech emotion analysis is crucial for advancing human-machine interaction.
• Current techniques for emotion recognition in speech require further development.
• Accurate speech emotion analysis has applications in customer service, lie detection, and feedback analysis.

Purpose of the Study:

• To propose a novel method for enhancing speech emotion analysis performance.
• To improve the accuracy and robustness of emotion recognition in speech signals.

Main Methods:

• The proposed method involves pre-processing, feature description using spectro-temporal modulation (STM) and entropy features, and feature extraction via Convolutional Neural Networks (CNN).
• Classification is performed using a combination of Gamma Classifier (GC) and Error-Correcting Output Codes (ECOC).
• The method was evaluated on the Berlin and ShEMO speech emotion datasets.

Main Results:

• The proposed method achieved an average accuracy of 93.33% on the Berlin dataset and 85.73% on the ShEMO dataset.
• Performance improvements of at least 6.67% were observed compared to existing methods.
• The combination of STM, entropy features, CNN, GC, and ECOC demonstrated high efficacy in speech emotion recognition.

Conclusions:

• The developed method offers a significant advancement in speech emotion analysis.
• The proposed approach provides a more accurate and effective solution for emotion recognition in human-machine interaction.
• This research contributes to the evolution of AI by enhancing its ability to understand human emotions from speech.