Medical speech display system using voice recognition software for patients with hearing disabilities

Abstract

We are currently studying what kinds of problems occur when using the software with the kind of vocabulary used between doctors and patients. Using the voice recognition software "Dragon Naturally Speaking 2005" (DNS for short) from Scan Soft, we input the words from everyday medical usage into word processing software (Microsoft "Word"). Specifically, we created a model conversation between doctors and patients. Using DNS we dictated these documents into "Word". We studied the correctness of the conversion of the words and phrases and the types of conversion errors. Based on the number of errors we established 4 grades of difficulty in conversion and analyzed the causes of misconversion. Among total 9464 words the 341 (3.6%) words and phrases of misconversion were classified as follows: 77.3% into grade 2. easy by training, 8.5% into grade 3. difficult and 14.2% into grade 4. very difficult. The main causes of difficulty in conversion were failure to properly recognize pronunciation, faulty ranking of homophones or too many listed homophones and words not being registered in the dictionary, coined words. In order to use the voice recognition system in everyday practice, the creation of a language dictionary specifically for medical communication may be needed.