A Novel User Emotional Interaction Design Model Using Long and Short-Term Memory Networks and Deep Learning

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Signal Processing

Background:

• Emotional design is crucial for enhancing user experience and emotional resonance in products.
• Accurate emotion recognition is vital for machines in emotional interaction design to understand user states.
• Deep learning offers a promising approach for more effective emotion recognition.

Purpose of the Study:

• To develop a deep learning mechanism for accurate and effective emotion recognition to optimize interactive systems.
• To explore user characteristics for machine emotion recognition, focusing on speech.
• To propose and validate a speech-based emotion recognition method.

Main Methods:

• Utilized Mel-Frequency Cepstral Coefficients (MFCC) as input for an improved Long and Short-Term Memory (ILSTM) network.
• Implemented peephole connections and unit state input in ILSTM for enhanced data integrity and accuracy.
• Applied a self-attention mechanism in the attention layer to weigh speech signal frames for emotion recognition.

Main Results:

• The proposed ILSTM model with an attention layer demonstrated effectiveness in speech emotion recognition.
• Experiments on EMO-DB and CASIA datasets validated the model's performance.
• The method successfully distinguished different emotions using weighted speech features.

Conclusions:

• The developed speech-based emotion recognition model is effective for optimizing interactive systems.
• The research validates the feasibility of emotional interaction system design using deep learning.
• Accurate emotion recognition through speech processing significantly improves user experience.