The use of speech based distant services from mobile devices requires sufficient network performance and disponsibility. Low rate speech coding reduces significantly these resource requirements, but it could have also a negative impact on speech quality. In this paper, we study the effect of Tetra transcoding on a speech recognition system. Experiments are conducted on an continuous speech large vocabulary task, using the LIA automatic speech recognition system (SPEERAL). We use the French broadcast news corpus provided for the Ester evaluation campaign. We first perform recognition using our baseline wide band system; results are then compared to one obtained using acoustic models trained on transcoded data. Our results show that processing of transcoded speech requires the adaptation of ASR system in order to reach a good level of performance.
Cite as: Georges, L., Pascal, N., Bertrand, R., Bonastre, J.-F. (2005) Automatic transcription of Tetra transcoded broadcast news. Proc. Biennial on DSP for In-Vehicle and Mobile Systems, paper A1-7
@inproceedings{georges05_dspinv,
author={Linares Georges and Nocera Pascal and Ravera Bertrand and Jean-François Bonastre},
title={{Automatic transcription of Tetra transcoded broadcast news}},
year=2005,
booktitle={Proc. Biennial on DSP for In-Vehicle and Mobile Systems},
pages={paper A1-7}
}