Speech intelligibility training using automatic speech recognition technology

Abstract

This paper describes two computer-based speech training systems that have been developed over the past decade of research. This research began at Indiana University and has been continued and commercialized at Communication Disorders Technology, Inc. The Indiana Speech Training Aid (ISTRA) uses a speaker-dependent speech recognizer and is designed for use by speech-language pathologists treating clients with articulation disorders of various sorts. The HearSay system employs a speaker-independent recognizer and is designed to improve the segment-level intelligibility of learners of English as a second language. Both systems employ minimal-pairs contrast training and share the underlying assumption that effective speech training requires intensive individual practice accompanied by evaluative feedback. The key technological achievements of ISTRA and HearSay are the methods employed for providing valid feedback on the intelligibility of the client’s speech. This paper will describe these methods and report the results of new training studies that demonstrate that computer-based speech training can be an effective method for improving the English speech intelligibility of learners of English as a second language. [Work supported by Grants No. DC00893 and No. DC02213 from the NIH–National Institute on Deafness and Other Communication Disorders.]