Knowledge Distillation for Small-footprint Highway Networks
This work addresses the need for efficient speech recognition models in resource-constrained environments, representing an incremental improvement over previous compact models.
The paper tackled the problem of deploying neural network models in embedded devices by using knowledge distillation to train a compact highway deep neural network (HDNN) for speech recognition, achieving significantly improved accuracy with less than 0.8 million parameters and narrowing the gap to a larger DNN with 30 million parameters.
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.