文件名称：Hybrid LSTM-FSMN Networks for Acoustic Modeling.pdf
文件大小：265KB
文件格式：PDF
更新时间：2022-08-17 05:03:43
他
Contextual information is key to training acoustic models. Recursive neural networks (RNNs) such as Long short term memory (LSTMs) are context-aware due to their recurrence mechanism, and usually outperform conventional neural networks. Connectionist temporal classification (CTC) allows neural 你networks to be trained to output the desired sequence of feature labels, but without the requirement to label specific feature vectors with specific labels.