Effects of channel conditions on the performance of ASR over GSM networks using full rate speech codec

Abstract

In this paper, the performance of Automatic Speech Recognition (ASR) and Mean Opinion Score (MOS) are evaluated under different channel conditions over GSM Network using standard narrow band Full Rate speech and channel coding algorithms. CMU SPHINX toolkit and TIMIT speech database (8 kHz raw files) are used to evaluate the ASR accuracy. MOS values are estimated, using PESQ P.862, as recommended by ITU-T (International Telecommunication Union). In this evaluation, Rayleigh distribution for generation of channel noise is used to simulate the channel conditions. The results obtained in the experiments are reported and analyzed.