Implementing machine learning techniques for continuous emotion prediction from uniformly segmented voice recordings.

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Speech Processing

Background:

• Emotional recognition from audio is crucial for AI development.
• Current methods face challenges with short audio samples and diverse datasets.

Purpose of the Study:

• To develop a novel method for accurate and efficient emotion recognition from short audio samples (1.5s).
• To improve AI's emotional intelligence for better human-computer interaction.

Main Methods:

• Utilized 1,510 audio samples in German and English.
• Employed Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and a hybrid C-DNN model.
• Extracted features for emotion prediction, addressing dataset heterogeneity and audio trimming complexities.

Main Results:

• Models significantly surpassed random guessing in accuracy.
• Performance aligned closely with human evaluative benchmarks.
• Demonstrated effectiveness in recognizing emotions from brief audio clips.

Conclusions:

• The proposed methodology shows significant potential for real-time emotion detection in continuous speech.
• Findings contribute to advancing AI's emotional intelligence and applications.
• Overcame challenges of diverse datasets and short audio sample integration.