The JD AI Speaker Verification System for the FFSVC 2020 Challenge

Keyphrases

• Speaker Recognition System 100%
• AI Speaker 100%
• Far-field Speaker Verification 100%
• Verification Challenges 100%
• Beamforming 50%
• Mandarin 25%
• Training Data 25%
• Training Set 25%
• Fusion Method 25%
• Factorized 25%
• U-Net 25%
• Channel Switching 25%
• Far-field Data 25%
• Delayed Neural Networks 25%
• Multiple Methods 25%
• Output Value 25%
• Equal Error Rate 25%
• Microphone Array 25%
• Self-attentive 25%
• Text-independent Speaker Verification 25%
• Field Text 25%
• Two-level Fusion 25%
• Voice Channels 25%
• Weighted Prediction Error 25%
• Single Microphone 25%
• Statistical Pooling 25%
• Speaker Characteristics 25%
• Network Transformers 25%
• Detection Cost Function 25%

Engineering

• Artificial Intelligence 100%
• Beamforming 100%
• Prediction Error 50%
• Field Data 50%
• Model Structure 50%
• Delay Time 50%
• Cost Function 50%
• Output Value 50%
• Equal Error Rate 50%
• Microphone Array 50%
• Text Field 50%
• Final Result 50%
• Resnet 50%

Computer Science

• Artificial Intelligence 100%
• Verification System 100%
• Speaker Verification 100%
• Training Data 16%
• Prediction Error 16%
• Multiple Method 16%
• time-delay 16%
• Structure Model 16%
• Residual Neural Network 16%
• Preprocessing 16%
• Equal Error Rate 16%
• Open Access 16%
• Transformer Neural Network 16%