A FAX Reader for the Blind

We describe an experiment in which a blind person places arbitrary printed pages in a FAX machine and soon afterwards hears the contents read out loud over the telephone by a synthesized voice. This study focuses on technical issues including the accuracy achievable by stateof-the-art optical character recognition operating on FAX images, methods to improve the intelligibility of synthesized speech for this application, and ease of interaction with the user. In a small-scale trial of the service under laboratory conditions, we have observed that accuracy and intelligibilit y on some commonlyoccurring types of documents, including typewritten letters, is usefully high. We believe that technology developments in the near future will support services acceptable to a wide range of users. 1. Introduct ion We descr ibe a small-sca le exploratory study of a new telecommunica tions ser vice conce pt directe d at visually-impaired populations, to rea d hardcopy printed text out loud over the telephone fully automatically. This re quires integrating state-of -the-a rt technologies for voice, data, and image proce ssing. The ser vice ac cepts images of documents using fa csimile transmission (FAX) , translates the image to ASCII text using optical char ac ter rec ognition (OCR), and conver ts the text to synthesized speec h (TT S). Due to the ubiqito usness of TouchTone telephones and FAX machines, such a servic e could enable visually-impaired people to cope with paper copies of documents and letters in a convenient and inexpensive manner , both at work and at home. We have car ried out a smallscale fe asibilit y trial under labora tory conditions, focusing on technica l issues including the acc urac y ac hievable by state-of -theart OCR opera ting on FAX images, methods to improve the intelli gibil ity of TTS in this application, and ease of interac tion with the user. The results of the study show that the ser vice conce pt is both technica lly fe asible and useful to the blind using curr ently-ava ilable technology. As a guide to future work, we propose a set of enhanc ements, principally in human-f actor s design and OCR ac cura cy, many of which should be straightforward to implement, and a fe w of which may re quire fur ther re sear ch. 24th Annual Asilomar Conference on Signals, Systems, and Computers, Pacific Grove, California, November 5-6, 1990.  1990 Maple Press