Random Deep Belief Networks for Recognizing Emotions from Speech Signals.

Area of Science:

• Artificial Intelligence
• Speech Processing
• Machine Learning

Background:

• Speech emotion recognition (SER) is crucial for human-computer interaction.
• Current SER methods face challenges with low recognition accuracy due to limited feature representation.
• Deep belief networks (DBN) offer hierarchical feature learning capabilities for speech signals.

Purpose of the Study:

• To enhance speech emotion recognition accuracy using an advanced machine learning approach.
• To leverage the representational power of deep belief networks for improved SER.
• To introduce a novel ensemble method for robust emotion detection from speech.

Main Methods:

• Extraction of low-level speech features.
• Construction of multiple random subspaces from extracted features.
• Application of deep belief networks (DBN) within each subspace for hierarchical feature learning.
• Ensemble of DBNs using majority voting for final emotion classification.

Main Results:

• The proposed random deep belief network (RDBN) ensemble method demonstrated superior performance.
• Experimental results on benchmark datasets showed higher accuracy compared to existing methods.
• The RDBN approach effectively addressed the limitations of feature representation in SER.

Conclusions:

• The ensemble of random deep belief networks (RDBN) is an effective method for improving speech emotion recognition.
• Hierarchical feature learning via DBNs in random subspaces enhances SER accuracy.
• This approach offers a promising direction for more accurate and reliable emotion detection from speech.