Deep Cross-Corpus Speech Emotion Recognition: Recent Advances and Perspectives

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Automatic speech emotion recognition (SER) is crucial for human-computer interaction (HCI).
• Current SER methods often fail in real-world applications due to performance limitations in cross-corpus and cross-language settings.
• Training and testing on single corpora/languages do not reflect practical diversity.

Purpose of the Study:

• To provide a comprehensive survey of cross-corpus SER techniques.
• To focus on advanced deep learning models for cross-corpus SER.
• To highlight challenges, opportunities, and future trends in the field.

Main Methods:

• Review of existing literature on cross-corpus SER.
• Analysis of deep learning techniques (supervised, unsupervised, semi-supervised) applied to cross-corpus SER.
• Identification of common challenges and emerging opportunities.

Main Results:

• Deep learning models show promise for improving cross-corpus SER performance.
• Significant performance gaps exist when applying SER models across different corpora or languages.
• The survey categorizes various deep learning approaches for cross-corpus SER.

Conclusions:

• Cross-corpus SER remains a challenging but vital area for HCI.
• Deep learning offers powerful solutions for adapting SER models to diverse data.
• Future research should address data diversity and model generalization for robust SER.