Neural net speech recognizers — Voice remote control devices for disabled people

Abstract

Reliable real world speech recognition is still a non-trivial task. One of the most challenging applications is voice control for disabled people. For many users of such a system, voice is the only means of communication — they are not able to activate or deactivate their voice control manually, or to correct false actions resulting from speech recognition errors. A voice remote control for disabled people is a 24-hours-a-day device which must respond reliably and, at the same time, must avoid unintentional activation deu to background noise, music or conversation. Moreover, in order to allow for emergency calls under all circumstances, it should not depend on external power supply — the hardware must be of low power consumption, thus requiring speech recognition algorithms of low computational complexity. With neural net speech recognizers, it was possible to build voice remote control devices that fulfill all these requirements simultaneously.