Feature enhancement by bidirectional LSTM networks for conversational speech recognition in highly non-stationary noise

Keyphrases

• Bidirectional Long Short-term Memory (BiLSTM) 100%
• Feature Enhancement 100%
• Non-stationary Noise 100%
• Conversational Speech Recognition 100%
• Speech Features 40%
• Spontaneous Speech 40%
• Recurrent Neural Network 20%
• Challenging Tasks 20%
• Recognition Task 20%
• Neural Network Architecture 20%
• Performance Gain 20%
• Contextual Information 20%
• Signal-to-noise Ratio 20%
• Memory-based 20%
• Low Signal-to-noise Ratio 20%
• Noisy Conditions 20%
• Recognition Performance 20%
• Temporal Context 20%
• Feature Generation 20%
• Non-negative Matrix Factorization 20%
• Multi-condition Training 20%
• Probabilistic Features 20%
• Conventional Neural Network 20%
• Clear Speech 20%
• Enhancement Approaches 20%
• Speech Enhancement 20%
• Word Accuracy 20%
• Bidirectional Long Short-term Memory Recurrent Neural Network 20%
• Regression Task 20%
• Ratio Level 20%
• Variable Noise 20%

Computer Science

• Speech Recognition 100%
• Bidirectional Long Short-Term Memory Network 100%
• Recurrent Neural Network 33%
• Noise-to-Signal Ratio 33%
• Spontaneous Speech 33%
• Neural Network Architecture 16%
• Performance Gain 16%
• Noisy Condition 16%
• Regression Task 16%
• Recognition Performance 16%
• Feature Generation 16%
• Temporal Context 16%
• nonnegative matrix factorization 16%
• Context Information 16%
• Conventional Neural Network 16%
• Speech Enhancement 16%