Pathologies not included: Survey of premillennium speech recognition software

An October 1999 survey of several leading speech recognition software manufacturers revealed that the speech technology industry is unprepared to serve users with vocal pathologies in 2000. Companies surveyed had neither prepared for consumers with vocal pathologies nor had immediate plans to alter systems to accommodate the pathological voice by 2002. Examples of vocal pathologies include leukoplakia, carcinoma, hyperfunction, polyps, and nodules. While the aforementioned pathologies involve direct impairment to the vocal mechanism (e.g., swollen pharyngeal mucous membranes, vocal fold polyps), neurological disorders such as Parkinsons disease also cause vocal pathologies (e.g., dysarthrias, spasmodic dysphonia). Some companies are beginning to incorporate natural language processing for consumers with aphasia; however, many of the top producers of speech recognition engines have not considered the pathological voice within their disability market. Acknowledging that the speech technology market is still new, speech recognition software developers note that lack of time and manpower prohibits concentration on groups with special needs. This research examines the feasibility of altering linguistic grammars of templates from healthy, prototypical voices to accommodate the phonetic changes that are observed in the pathological voice. Backus‐Naur form and Chomskys syntactic structures are used to modify existing grammar templates.