Voice recognition performance with naive versus practiced speakers

Abstract

The purpose of the current study was to determine the accuracy of a current voice recognition device (VRD) when used by naive speakers versus practiced speakers, in a speaker independent mode (one in which the VRD device relies on the speech patterns of individuals other than the current speaker). It is conceivable that in future applications of VR technology, it may be costly or impractical to provide practice and training to all users. The findings suggest that first time users of VR equipment, will obtain 96.85% recognition accuracy, a level at least as high as that obtained by users who have received training or practiced speaking to the VRD. Neither nonrecognitions (e.g., errors where the system rejects the input and responds, in effect, with I don't understand you, say it again) or misrecognitions (e.g., errors where the system accepts the input but mistakes it for a different input) differed significantly for naive speakers versus practiced speakers. Furthermore, the misrecognition rate for naive speakers was only 1.11%. It was concluded that training and practice may not always be necessary in order to obtain optimum performance in the human-VRD system. Without the need for practice, which implies modifying the human's behavior, the human-machine interaction is more natural, the friendliness of the VRD is enhanced, and the cost of the VR system use is reduced