Longitudinal study of voice recognition in children

Abstract

Speaker recognition as a biometric modality is on the rise in the consumer marketplace for banking, online services, and personal assistant services with a potential for wider application areas. Most current applications involve adults. One of the biggest challenges in speaker recognition for children is the change in the voice properties as a child age. This work proposes a baseline longitudinal dataset from the same 30 children in the age group of 4 to 14 years over a time frame of 2.5 years and evaluates speaker recognition performance in children with the available speaker recognition technology.