Human-Computer Interaction for Recognizing Speech Emotions Using Multilayer Perceptron Classifier.

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Speech Processing

Background:

• Human-computer interaction (HCI) is shifting towards intuitive modalities like voice control.
• Speech contains rich information beyond words, including speaker emotions, mood, and intent.
• Accurate speech emotion recognition is crucial for advanced HCI systems, but integrating findings across disciplines remains challenging.

Purpose of the Study:

• To develop and evaluate an artificial intelligence model for detecting emotions in human speech.
• To leverage deep learning methods for real-time emotion analysis from voice data.
• To classify eight distinct emotional states from vocalizations.

Main Methods:

• Utilized the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) dataset, comprising over 2000 speech and song recordings from 24 actors.
• Trained a multilayer perceptron (MLP) classifier, a supervised learning algorithm, for emotion classification.
• Evaluated model performance against similar studies and benchmarked results.

Main Results:

• The proposed model achieved an overall accuracy of 81% in classifying eight different emotion classes.
• Successfully detected emotions including neutral, calm, happy, sad, angry, fearful, disgusted, and surprised moods.
• Demonstrated the feasibility of real-time emotion analysis in live speech using AI.

Conclusions:

• Artificial intelligence, particularly deep learning, shows significant promise for accurate speech emotion recognition.
• The developed MLP model provides a robust method for classifying multiple emotions from voice.
• This research contributes to more sophisticated and emotionally aware human-computer interaction systems.