Mutual Information Enhanced Training for Speaker Embedding

Keyphrases

• Deep InfoMax 100%
• Detrimental Effects 20%
• Enhanced Training 100%
• Frame-level Features 60%
• Global Pooling 20%
• Imbalance Problem 20%
• Low-level Features 20%
• Meaningful Information 20%
• Mutual Information 100%
• Mutual Information Estimation 60%
• Mutual Information Maximization 20%
• Performance Gain 40%
• Pooling Method 20%
• Representation Learning 20%
• Self-supervised Learning 20%
• Speaker Embedding 100%
• Speaker Verification 20%
• Unsupervised Learning 20%
• VoxCeleb 20%

Computer Science

• Dimensional Space 16%
• Information Loss 16%
• Mutual Information 100%
• Performance Gain 33%
• Representation Learning 16%
• Speaker Verification 16%