Exploring Intrinsic Speaker Qualities Via an Analysis of Automatic Speaker Recognition System Errors
Lara Lynn Stoll and Nelson Morgan
Although a great deal of progress has been made in the task of automatic speaker recognition, there are still many challenges that remain. A relatively limited amount of work has been done to find or characterize speakers who may be inherently hard to recognize. We perform an error analysis of automatic speaker recognition systems, with a focus on identifying the characteristics of speakers who cause a disproportionately large number of the errors. By considering a range of intrinsic speaker qualities, including physical attributes, prosodic characteristics, and accents or dialects, we aim to find shared attributes among the speakers who cause the most recognition errors. Ultimately, the results of such an error analysis will be used to improve speaker recognition performance.