Classification of Infant Crying Sounds Using SE-ResNet-Transformer.

Area of Science:

• Artificial Intelligence
• Speech Emotion Analysis
• Infant Communication

Background:

• Speech emotion analysis is crucial for AI understanding human communication.
• Infant crying is a primary method of emotional expression for babies.
• Cries convey vital information such as hunger, pain, and discomfort.

Purpose of the Study:

• To develop an advanced classification model for analyzing infant cry emotions.
• To improve the accuracy and efficiency of infant emotion detection systems.

Main Methods:

• Utilized a hybrid ResNet and transformer model architecture.
• Employed feature engineering on Mel-frequency cepstral coefficient (MFCC) features from infant cries.
• Integrated SE attention mechanism modules within residual blocks to optimize channel weights.

Main Results:

• Achieved a high accuracy rate of 93% in infant cry classification experiments.
• Demonstrated significantly shorter training times compared to traditional models.
• Showcased superior accuracy over existing methods for infant emotion analysis.

Conclusions:

• The proposed ResNet-transformer model offers an efficient and stable solution for infant cry classification.
• This AI-driven approach enhances the ability to interpret infant emotional states.
• The findings contribute to advancements in affective computing and infant care technology.