Rating of Speech Quality in Mobile Phone Networks

Abstract

Mobile phone users rely only on their intuition to evaluate the quality of service provided by their network operator. In this paper, we propose an output-based objective method to rate the speech quality in mobile phones. This rating, when averaged over time, will be useful in comparing between different mobile phone networks. Moreover, the rating could be incorporated into the calls tariff of mobile phones.Here, a time-delay multilayer neural network is used to rate the speech quality after a proper training stage. The training input set consists of speech features (such as linear predictive coefficients LPC and per- frame energy). The training target is a customized per-frame speech quality measure. We performed extensive simulations using the proposed method. We use speech samples from simulated channels, real GSM channel and from the more elaborate CTIMIT database. These simulations show that the proposed method can predict the speech quality within reasonable accuracy. Keywords: Speech quality, mobile network, CTIMIT database, time-delay neural network