Effects of equipment variations on speaker recognition error rates.

摘要

Scope and Method of Study. The purpose of this study was to examine the effects that equipment variation has on speaker recognition performance. Specifically microphone variation is investigated. The study examines the error rates of a speaker recognition system when microphones vary between the enrollment and testing phases. The study also examines the error rates of a speaker recognition system when microphones differ in similar environments and conditions. The metric for evaluation of effect is the false identity acceptance and the false identity rejection error rates.;Findings and Conclusions. The results of the research demonstrate that microphone variation has a major effect on speaker recognition error rates. Error rates include the rates of false acceptance and false rejection of identity. The effect of a training/enrollment microphone mismatched was significant. Mismatched conditions produce significantly greater false accept and false reject rates as compared to matched conditions. In fact, the mismatched condition had a more significant impact on error rates than noise. The research also demonstrates that speaker recognition error rates are microphone dependant. The microphone dependency was seen in both the matched and mismatched condition. However, the microphone dependency was more prevalent in the matched condition. Microphone selection has an effect on error rates in different environments and in matched and mismatched conditions. The research provides a method to evaluate the direct effects of microphone selection on speaker recognition systems.