Measurement of speaker recognizability

Speech processing algorithms have been evaluated with respect to intelligibility and quality. Many years of testing have established standard tests, namely the DRT for intelligibility and DAM for quality, but no such tests have been established for measuring speaker recognizability. This paper discusses the correlation between error rates of an automatic speaker identification algorithm with those of a human panel. In addition, it presents the correlation between error rates of the automatic speaker identification system and DAM scores.<>