The simulation of realistic acoustic input scenarios for speech recognition systems

A tool for simulating the acoustic conditions during the speech input to a recognition system and the transmission in telephone networks is presented in this paper. The simulation covers the hands-free speech input in rooms and the existence of noise in the background. Furthermore the presence of telephone frequency characteristics can be simulated. Finally the transmission in a cellular telephone system like GSM or UMTS is covered including the encoding and decoding of speech and the transmission over the erroneous radio channel. The tool has been realized by integrating functions from the ITU software library for implementing telephone frequency characteristics and the estimation of the speech level as well as software modules from ETSI and 3GPP for the AMR encoding and decoding of speech. A Web interface has been designed to experience the simulation tool with acoustic examples.