Evaluation of Three Distance Measures for Speaker-Independent Recognition in Telecommunication Services

The performance of automatic speech recognition (ASR) algorithms has matured sufficiently such that the prospect of telecommunication services incorporating ASR features is no longer considered remote. Unfortunately, current ASR systems developed to support these services are restricted by the amount of processing power available, the memory limitations on the number of recognition templates, and the financial constraints on the design of the ASR system. These restrictions have and, for some time, will prevent implementation of the more sophisticated ASR algorithms that are being developed. Therefore, to obtain the optimal performance, it will be necessary to tune various features of the standard ASR algorithms in order to satisfy service objectives.