Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition

TLDR

Deep belief network pre‑training robustly initializes deep neural networks, improving optimization and reducing generalization error. The authors propose a novel context‑dependent DNN‑HMM model for large‑vocabulary speech recognition that leverages deep belief networks and detail its components, application procedure, and modeling choices. The model is a pre‑trained DNN‑HMM hybrid that outputs a distribution over senones, trained using deep belief network pre‑training. On a business search dataset, the CD‑DNN‑HMM achieves 5.8–9.2% absolute sentence‑accuracy gains (16–23% relative error reduction) over conventional CD‑GMM‑HMMs trained with MPE or ML.

Abstract

We propose a novel context-dependent (CD) model for large-vocabulary speech recognition (LVSR) that leverages recent advances in using deep belief networks for phone recognition. We describe a pre-trained deep neural network hidden Markov model (DNN-HMM) hybrid architecture that trains the DNN to produce a distribution over senones (tied triphone states) as its output. The deep belief network pre-training algorithm is a robust and often helpful way to initialize deep neural networks generatively that can aid in optimization and reduce generalization error. We illustrate the key components of our model, describe the procedure for applying CD-DNN-HMMs to LVSR, and analyze the effects of various modeling choices on performance. Experiments on a challenging business search dataset demonstrate that CD-DNN-HMMs can significantly outperform the conventional context-dependent Gaussian mixture model (GMM)-HMMs, with an absolute sentence accuracy improvement of 5.8% and 9.2% (or relative error reduction of 16.0% and 23.2%) over the CD-GMM-HMMs trained using the minimum phone error rate (MPE) and maximum-likelihood (ML) criteria, respectively.