Low-Latency Sequence-to-Sequence Speech Recognition and Translation by Partial Hypothesis Selection

Keyphrases

• Speech Recognition 100%
• Low Latency 100%
• Speech Translation 100%
• Hypothesis Selection 100%
• Latency 50%
• Encoder-decoder 50%
• Off-line System 50%
• Word Error Rate 50%
• Reduced Latency 50%
• Selection Strategy 25%
• Reduction Method 25%
• Encoder 25%
• Original Model 25%
• Quality Metrics 25%
• Sequence Task 25%
• Generic Architecture 25%
• Transformer 25%
• Sequence-to-sequence 25%
• Latency Reduction 25%
• Partial Sequence 25%
• Incremental Inference 25%
• BLEU Score 25%
• How2 Dataset 25%
• Adaptation Procedure 25%
• Chunk-based 25%

INIS

• speech 100%
• hypothesis 100%
• comparative evaluations 50%
• errors 50%
• datasets 50%
• reduction 25%
• metrics 25%
• losses 25%
• accuracy 25%
• architecture 25%
• calculation methods 25%
• transcription 25%
• transformers 25%

Computer Science

• Speech Recognition 100%
• Word Error Rate 100%
• Seq2Seq 100%
• Use Case 50%
• Selection Strategy 50%
• Partial Sequence 50%
• Latency Reduction 50%
• Generic Architecture 50%