Deep-Net: A Lightweight CNN-Based Speech Emotion Recognition System Using Deep Frequency Features.

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Speech emotion recognition (SER) systems aim to identify human emotions from speech signals.
• Current SER methods face challenges with low accuracy across different languages, emotions, and datasets.
• Analyzing speech signals for emotion is complex due to varying frequencies and features.

Purpose of the Study:

• To propose a novel, lightweight, and effective SER model with low computational complexity and high recognition accuracy.
• To enhance the discriminative power of frequency features for improved emotion recognition.
• To develop an AI-driven system capable of efficiently recognizing emotions in speech.

Main Methods:

• Utilized a convolutional neural network (CNN) approach for SER.
• Employed a plain rectangular filter with a modified pooling strategy to learn deep frequency features.
• Trained the CNN model on extracted frequency features from speech data.

Main Results:

• The proposed CNN-based SER model achieved high recognition accuracy on benchmark datasets.
• Achieved 77.01% accuracy on the IEMOCAP dataset and 92.02% on the EMO-DB dataset.
• Demonstrated superior performance compared to existing state-of-the-art SER systems.

Conclusions:

• The developed lightweight CNN model offers a promising solution for accurate speech emotion recognition.
• The modified pooling strategy enhances the model's ability to discern emotional nuances in speech.
• This AI-driven approach represents a significant advancement in the field of SER.