A Comparison of Human and Machine Estimation of Speaker Age

Abstract

Abstract. The estimation of the age of a speaker from his or her voice has both forensic and commercial applications. Previous studies have shown that human listeners are able to estimate the age of a speaker to within 10 years on average, while recent machine age estimation systems seem to show superior performance with average errors as low as 6 years. However the machine studies have used highly non-uniform test sets, for which knowledge of the age distribution offers considerable advantage to the system. In this study we compare human and machine performance on the same test data chosen to be uniformly distributed in age. We show that in this case human and machine accuracy is more similar with average errors of 9.8 and 8.6 years respectively, although if panels of listeners are consulted, human accuracy can be improved to a value closer to 7.5 years. Both human and machines have difficulty in accurately predicting the ages of older speakers.Keywords: speaker profiling, speaker age prediction, computational paralinguistics. IntroductionThe estimation of the age of a speaker from an analysis of his or her voice has forensic applications -for example the profiling of perpetrators of crimes [1], commercial applications -for example targeted advertising, and technological applicationsfor example adaptation of a speech recognition system to a speaker [2].Many previous studies have looked at the performance of both human listeners and machine-learning systems for the estimation of age from speech. Unfortunately, variations in the data set, task and performance metric make these studies hard to compare. In our work we take the view that the natural task should be numerical estimation of the age of the speaker, and the natural performance metric should be the mean absolute error (MAE) of estimation. The MAE answers the question "how close is the average estimate to the actual age?" A recent review of previous studies on human listener judgments of speaker age may be found in [3]. Of the studies reported which used numerical age estimation and