Recently, convolutional neural networks (CNNs) have been shown to outperform the standard fully connected deep neural networks within the hybrid deep neural network / hidden Markov model (DNN/HMM) framework on the phone recognition task. In this paper, we extend the earlier basic form of the CNN and explore it in multiple ways. We first investigate several CNN architectures, including full and limited weight sharing, convolution along frequency and time axes, and stacking of several convolution layers. We then develop a novel weighted softmax pooling layer so that the size in the pooling layer can be automatically learned. Further, we evaluate the effect of CNN pretraining, which is achieved by using a convolutional version of the RBM. We show that all CNN architectures we have investigated outperform the earlier basic form of the DNN on both the phone recognition and large vocabulary speech recognition tasks. The architecture with limited weight sharing provides additional gains over the full weight sharing architecture. The softmax pooling layer performs as well as the best CNN with the manually tuned fixed-pooling size, and has a potential for further improvement. Finally, we show that CNN pretraining produces significantly better results on a large vocabulary speech recognition task.
Bibliographic reference. Abdel-Hamid, Ossama / Deng, Li / Yu, Dong (2013): "Exploring convolutional neural network structures and optimization techniques for speech recognition", In INTERSPEECH-2013, 3366-3370.