INTapt: Information-Theoretic Adversarial Prompt Tuning for Enhanced Non-Native Speech Recognition

Keyphrases

• Information Theory 100%
• Non-native Speech Perception 100%
• Prompt Tuning 100%
• Concatenated 60%
• L2 English 40%
• Training Model 40%
• L1 English 40%
• English Accents 40%
• Non-native 20%
• Feature Similarity 20%
• Automatic Speech Recognition 20%
• Speech Recognition System 20%
• Speech Corpus 20%
• Representation Bias 20%
• English Speech 20%
• Speech Recognition Performance 20%
• Speech Model 20%
• Model Weight 20%
• Feature Dependency 20%
• Adversarial Training 20%
• Self-supervised Speech Representations 20%
• CTC Loss 20%
• Pre-trained Speech Models 20%
• Pre-trained Based 20%

Computer Science

• Speech Recognition 100%
• Prompt Tuning 100%
• Pre-Trained Model 60%
• Representation Learning 20%
• Experimental Result 20%
• Recognition Performance 20%
• Speech Recognition System 20%
• speech corpus 20%
• Adversarial Machine Learning 20%