Enhanced Direct Speech-to-Speech Translation Using Self-supervised Pre-training and Data Augmentation

Keyphrases

• Data Augmentation 100%
• Self-supervised Pre-training 100%
• Speech-to-speech Translation 100%
• Machine Translation 66%
• S2ST 66%
• English Translation 33%
• Decoder 33%
• Discrete Units 33%
• Training Data 33%
• Automatic Speech Recognition 33%
• Spanish-English Bilinguals 33%
• Multi-task Learning 33%
• Target Speech 33%
• Weakly Supervised Training 33%
• Cascade System 33%
• Data Scarcity 33%
• Text-to-speech Synthesis 33%
• Model Evaluation 33%
• Speech Encoders 33%
• Speech Dataset 33%
• Discrete Representation 33%

Computer Science

• Data Augmentation 100%
• Machine Translation 66%
• Text To Speech 66%
• Speech Recognition 33%
• Performance Model 33%
• Multitask Learning 33%
• Speech Synthesis 33%
• Training Data 33%
• English Translation 33%