Adaptive Data Boosting Technique for Robust Personalized Speech Emotion in Emotionally-Imbalanced Small-Sample...

Area of Science:

• Artificial Intelligence
• Speech Processing
• Machine Learning

Background:

• Personalized emotion recognition aims to improve accuracy by training models on individual user data, unlike general models.
• Existing methods face a 'cold-start' problem, requiring extensive, balanced emotional speech data from target users, which is difficult to obtain in real-world scenarios.

Purpose of the Study:

• To address the cold-start problem in personalized speech emotion recognition.
• To develop a robust framework that enables effective personalized emotion recognition with limited and imbalanced user data.

Main Methods:

• Proposing a Robust Personalized Emotion Recognition Framework utilizing an Adaptive Data Boosting Algorithm.
• Incrementally building customized models by combining target user speech with other users' data.
• Employing SMOTE (Synthetic Minority Over-sampling Technique)-based data augmentation to enhance the dataset.

Main Results:

• The proposed framework demonstrates adaptability with small target user datasets.
• The method is effective in emotionally imbalanced data environments.
• Experiments using the IEMOCAP (Interactive Emotional Dyadic Motion Capture) database validated the approach.

Conclusions:

• The developed framework successfully mitigates the cold-start problem in personalized emotion recognition.
• This approach offers a practical solution for real-world applications requiring accurate, individualized emotion recognition systems.
• The adaptive data boosting and augmentation techniques enhance model performance with limited data.