Toward Voice Applications for Children

Abstract

This paper explores technology available for prototyping and product development of child speech recognition applications, and factors relevant for eventual delivery of good user experience. Since child voices are known to be more challenging than adults, and less commercial investment has been dedicated to child voice modeling, it is significant that recognition accuracy for very constrained distinct vocabulary grammars is encouragingly high using out-of-the box technology to recognize suburban varieties of English. While accuracy on more demanding recognition tasks seems insufficient for immediate product development without focused support by a research laboratory, it is high enough to suggest that the class of feasible child applications may soon widen as new resources for child recognition become available. Preliminary results reported here contribute to a roadmap in pursuit of broad demographic coverage for a widening set of child applications.