eduzhai > Applied Sciences > Engineering >

Sequential Routing Framework Fully Capsule Network-based Speech Recognition

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 42 pages

Abstract: Capsule networks (CapsNets) have recently gotten attention as a novel neuralarchitecture. This paper presents the sequential routing framework which webelieve is the first method to adapt a CapsNet-only structure tosequence-to-sequence recognition. Input sequences are capsulized then sliced bya window size. Each slice is classified to a label at the corresponding timethrough iterative routing mechanisms. Afterwards, losses are computed byconnectionist temporal classification (CTC). During routing, the requirednumber of parameters can be controlled by the window size regardless of thelength of sequences by sharing learnable weights across the slices. Weadditionally propose a sequential dynamic routing algorithm to replacetraditional dynamic routing. The proposed technique can minimize decoding speeddegradation caused by the routing iterations since it can operate in anon-iterative manner without dropping accuracy. The method achieves a 1.1 lower word error rate at 16.9 on the Wall Street Journal corpus compared tobidirectional long short-term memory-based CTC networks. On the TIMIT corpus,it attains a 0.7 lower phone error rate at 17.5 compared to convolutionalneural network-based CTC networks (Zhang et al., 2016).

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×