Speaker verification routines for ISDN and UPT access and security using artificial neural networks and time encoded speech (TES) data

摘要

An initial investigation into the use of time encoded speech (TES) fixed size, fixed dimension A-matrices as the input layer of a simple fast artificial neural network (FANN) configured to identify the acoustic output of ten cooperative male imposters each articulating twenty versions of the same phrase produced very favourable indications which led to the work described in the paper. It would appear that the TES data structures involved may be well matched to the demands of artificial neural network architectures when these are applied to some of the important time-sequence learning requirements inherent in the significant task of speaker verification. Such a combination may well enable these very powerful FANN analysis and classification tools to be applied to advantage to the spoken utterances of voice network users under realistic operational conditions.