Gender-Driven English Speech Emotion Recognition with Genetic Algorithm.

Area of Science:

• Computer Science
• Artificial Intelligence
• Speech Processing

Background:

• Accurate speech emotion recognition (SER) is crucial for personalized human-computer interaction.
• Gender-specific SER enhances empathy in technology, healthcare, and psychology.
• Existing models often struggle with nuanced emotional distinctions and feature selection.

Purpose of the Study:

• To develop a novel gender-emotion recognition model for improved speech analysis.
• To enhance feature selection and recognition accuracy using an optimized genetic algorithm (GA).
• To evaluate the model's performance against state-of-the-art algorithms.

Main Methods:

• Extraction of gender and emotion features from voice signals.
• Application of a genetic algorithm (GA) for high-dimensional feature processing and selection.
• Improvement of GA with novel crossover and mutation methods based on Fisher score feature ranking.
• Comparison with existing algorithms using Support Vector Machines (SVM) on four English datasets.

Main Results:

• The proposed GA-based model demonstrated superior performance across accuracy, precision, recall, F1-score, feature selection, and running time.
• Mel frequency cepstral coefficients (MFCC) and log MFCC were identified as key features for gender differentiation.
• Challenges remain in distinguishing between neutral, sad, and fearful emotions due to subtle vocal variations.

Conclusions:

• The novel GA-based approach significantly improves gender-based speech emotion recognition.
• Feature importance-driven GA optimization enhances model efficiency and accuracy.
• Further research is needed to address subtle emotion classification challenges.