语音识别LAS结构where d and y, are MLP networks. After training, the a; distribution Table 1: WER comparison on the clean and noisy Google voice
is typically very sharp and focuses on only a few frames of h; ci car
search task. The CLDNN-hMM system is the s
