Speaking rate estimation based on deep neural networks

Abstract

In this paper we propose a method for estimating speaking rate by means of Deep Neural Networks (DNN). The proposed approach is used for speaking rate adaptation of an automatic speech recognition system. The adaptation is performed by changing step in front-end feature processing according to the estimations of speaking rate. Experiments show that adaptation results using the proposed DNN-based speaking rate estimator are better than the results of adaptation using the speaking rate estimator based on the recognition results.