ABSTRACT
This chapter describes the design features of speech training systems that we and our colleagues have developed over the course of more than a decade of research. This research began at Indiana University in 1986 and, since 1989, has been continued at Communication Disorders Technology, Inc. Each of the systems described here was developed to serve the training needs of rather different sets of clients. All three systems employ automatic speech recognition (ASR) technology for the purpose of providing feedback to clients about the quality or intelligibility of their speech. It is a basic cognitive assumption underlying all this work that effective training of speech and language structures for persons with disorders of various types, or for second language learners, requires a large amount of individual drill with valid feedback. To the extent that they provide valid feedback about the intelligibility of an utterance, computer-based training systems employing ASR afford amounts of individual training not available in the typical clinical or classroom setting. The first training system is the Indiana Speech Training Aid (ISTRA; Kewley-Port, Watson, Elbert, Maki, & Reed, 1991). This system is designed for use by speech-language pathologists in the treatment of the speech production deficits of normal-hearing, misarticulating, or hearing-impaired children. It uses a speaker-dependent recognizer that provides client feedback based on a numerical score that has been shown to be well correlated with human judgments of speech quality (Watson, Reed, KewleyPort, & Maki, 1989). The ISTRA system trains clients on their productions of whole words and short phrases.
