Knowledge Distillation for Small-footprint Highway Networks

Deep learning has significantly advanced state-of-the-art of speech recognition in the past few years. However, compared to conventional Gaussian mixture acoustic models, neural network models are usually much larger, and are therefore not very deployable in embedded devices. Previously, we investigated a compact highway deep neural network (HDNN) for acoustic modelling, which is a type of depth-gated feedforward neural network. We have shown that HDNN-based acoustic models can achieve comparable recognition accuracy with much smaller number of model parameters compared to plain deep neural network (DNN) acoustic models. In this paper, we push the boundary further by leveraging on the knowledge distillation technique that is also known as {\it teacher-student} training, i.e., we train the compact HDNN model with the supervision of a high accuracy cumbersome model. Furthermore, we also investigate sequence training and adaptation in the context of teacher-student training. Our experiments were performed on the AMI meeting speech recognition corpus. With this technique, we significantly improved the recognition accuracy of the HDNN acoustic model with less than 0.8 million parameters, and narrowed the gap between this model and the plain DNN with 30 million parameters.

Segmental Recurrent Neural Networks for End-to-end Speech Recognition

We study the segmental recurrent neural network for end-to-end acoustic modelling. This model connects the segmental conditional random field (CRF) with a recurrent neural network (RNN) used for feature extraction. Compared to most previous CRF-based acoustic models, it does not rely on an external system to provide features or segmentation boundaries. Instead, this model marginalises out all the possible segmentations, and features are extracted from the RNN trained together with the segmental CRF. In essence, this model is self-contained and can be trained end-to-end. In this paper, we discuss practical training and decoding issues as well as the method to speed up the training in the context of speech recognition. We performed experiments on the TIMIT dataset. We achieved 17.3 phone error rate (PER) from the first-pass decoding --- the best reported result using CRFs, despite the fact that we only used a zeroth-order CRF and without using any language model.

Top-down Tree Long Short-Term Memory Networks

Long Short-Term Memory (LSTM) networks, a type of recurrent neural network with a more complex computational unit, have been successfully applied to a variety of sequence modeling tasks. In this paper we develop Tree Long Short-Term Memory (TreeLSTM), a neural network model based on LSTM, which is designed to predict a tree rather than a linear sequence. TreeLSTM defines the probability of a sentence by estimating the generation probability of its dependency tree. At each time step, a node is generated based on the representation of the generated sub-tree. We further enhance the modeling power of TreeLSTM by explicitly representing the correlations between left and right dependents. Application of our model to the MSR sentence completion challenge achieves results beyond the current state of the art. We also report results on dependency parsing reranking achieving competitive performance.

Tied Probabilistic Linear Discriminant Analysis for Speech Recognition

Acoustic models using probabilistic linear discriminant analysis (PLDA) capture the correlations within feature vectors using subspaces which do not vastly expand the model. This allows high dimensional and correlated feature spaces to be used, without requiring the estimation of multiple high dimension covariance matrices. In this letter we extend the recently presented PLDA mixture model for speech recognition through a tied PLDA approach, which is better able to control the model size to avoid overfitting. We carried out experiments using the Switchboard corpus, with both mel frequency cepstral coefficient features and bottleneck feature derived from a deep neural network. Reductions in word error rate were obtained by using tied PLDA, compared with the PLDA mixture model, subspace Gaussian mixture models, and deep neural networks.