Very Deep Convolutional Neural Networks for Robust Speech Recognition

Abstract

This paper describes the extension and optimization of our previous work on very deep convolutional neural networks (CNNs) for effective recognition of noisy speech in the Aurora 4 task. The appropriate number of convolutional layers, the sizes of the filters, pooling operations and input feature maps are all modified: the filter and pooling sizes are reduced and dimensions of input feature maps are extended to allow adding more convolutional layers. Furthermore appropriate input padding and input feature map selection strategies are developed. In addition, an adaptation framework using joint training of very deep CNN with auxiliary features i-vector and fMLLR features is developed. These modifications give substantial word error rate reductions over the standard CNN used as baseline. Finally the very deep CNN is combined with an LSTM-RNN acoustic model and it is shown that state-level weighted log likelihood score combination in a joint acoustic model decoding scheme is very effective. On the Aurora 4 task, the very deep CNN achieves a WER of 8.81%, further 7.99% with auxiliary feature joint training, and 7.09% with LSTM-RNN joint decoding.

accepted by SLT 2016

References in corpus (3)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Stochastic Pooling for Regularization of Deep Convolutional Neural Networks
• Learning Hidden Unit Contributions for Unsupervised Acoustic Model Adaptation

Cited by in corpus (5)

• Recent Advances in Convolutional Neural Networks
• Residual Convolutional CTC Networks for Automatic Speech Recognition
• Physical Accuracy of Deep Neural Networks for 2D and 3D Multi-Mineral Segmentation of Rock micro-CT Images
• ML-LBM: Machine Learning Aided Flow Simulation in Porous Media
• An Exploration of Mimic Architectures for Residual Network Based Spectral Mapping