Adaptation Of Context-Dependent Deep Neural Networks For Automatic Speech Recognition

In this paper, we evaluate the effectiveness of adaptation methods for context-dependent deep-neural-network hidden Markov models (CD-DNN-HMMs) for automatic speech recognition. We investigate the affine transformation and several of its variants for adapting the top hidden layer. We compare the affine transformations against direct adaptation of the softmax layer weights. The feature-space discriminative linear regression (fDLR) method with the affine transformations on the input layer is also evaluated. On a large vocabulary speech recognition task, a stochastic gradient ascent implementation of the fDLR and the top hidden layer adaptation is shown to reduce word error rates (WERs) by 17% and 14%, respectively, compared to the baseline DNN performances. With a batch update implementation, the softmax layer adaptation technique reduces WERs by 10%. We observe that using bias shift performs as well as doing scaling plus bias shift

0000366.pdf
PDF file

In  SLT 2012

Details

TypeInproceedings
> Publications > Adaptation Of Context-Dependent Deep Neural Networks For Automatic Speech Recognition