CUED Publications database

Sequence training of DNN acoustic models with natural gradient

Haider, A and Woodland, PC (2017) Sequence training of DNN acoustic models with natural gradient. In: UNSPECIFIED pp. 178-184..

Full text not available from this repository.


Deep Neural Network (DNN) acoustic models often use discriminative sequence training that optimises an objective function that better approximates the word error rate (WER) than frame-based training. Sequence training is normally implemented using Stochastic Gradient Descent (SGD) or Hessian Free (HF) training. This paper proposes an alternative batch style optimisation framework that employs a Natural Gradient (NG) approach to traverse through the parameter space. By correcting the gradient according to the local curvature of the KL-divergence, the NG optimisation process converges more quickly than HF. Furthermore, the proposed NG approach can be applied to any sequence discriminative training criterion. The efficacy of the NG method is shown using experiments on a Multi-Genre Broadcast (MGB) transcription task that demonstrates both the computational efficiency and the accuracy of the resulting DNN models.

Item Type: Conference or Workshop Item (UNSPECIFIED)
Divisions: Div F > Machine Intelligence
Depositing User: Cron Job
Date Deposited: 02 Mar 2019 20:05
Last Modified: 13 Apr 2021 09:33
DOI: 10.1109/ASRU.2017.8268933