Takayuki Watanabe

A prototype lunch delivery Web system for the visually impaired was developed based on the analysis of human (the visually impaired who wants to order a lunch) to human (the sighted who helps the visually impaired to decide lunch by reading aloud lunch menu) dialog. Firstly, human-to-human dialog was analyzed qualitatively to find important points of the dialog. Secondly, the dialog was analyzed in terms of the intension of a speaker. Based on these analysis, the prototype system was developed, which consists of three steps: 1) rough selection (candidate items are roughly selected based on categories), 2) selection of favorites (favorite items are selected from candidate items and stored in the system), and 3) final selection (one item is selected with detailed information for each item). The evaluation of the prototype system by the visually impaired shows that 1) the user can select a lunch item with this system, 2) the user said the system is better than human-to-human dialog because he could easily retrieve past information, i.e. navigate back in a page, 3) the current Web page is easy to understand, 4) "go back" function is necessary, and 5) the user prefers XHTML Web site rather than VoiceXML site.

The objective of the current study is to determine important facts in describing images or photos with speech without using the sense of sight. We prepared 10 types of descriptions for 9 photographs respectively. The subjects were required to evaluate each description in terms of "easy to understand" and other 7 points. The result shows that "easy to understand" consists of, "good tones", "simple words", "clearness", "enough explanation", and "shortness" in this order. We also interviewed people with visual disabilities to ask what kinds of description was easy to understand.

Web content accessibility guidelines such as JIS X 8341-3 and WCAG 2.0 working draft require the separation of content structure from presentation. In (X)HTML, content structure is marked up with heading elements, list elements, etc. Objective evaluation of how content-presentation separation improves accessibility is needed to force Web authors to follow these accessibility guidelines. The current paper measures how usability and accessibility is improved when content is marked up with heading elements. The experiment was carried out firstly with four sighted persons and found that structured Web pages improve task completion time by 50% to 90%. The second experiment was carried out with one visually impaired and found that structured Web pages improve task completion time by 30% to 70%.

A text-to-speech synthesizer that would produce easily understandable voices at very fast speaking rates is expected to help persons with visual disability to acquire information effectively with screen reading software. We investigated the intelligibility of Japanese text-to-speech systems at very fast speaking rates using four-digit random numbers as the vocabulary of the recall test. As a result of the experiments with young subjects and elderly subjects, the significant learning effects were not observed in case of the elderly subjects as compared with the young subjects.

An experiment of performing human-to-human dialogs to assist visually-impaired people was conducted for acquiring knowledge required for developing dialog systems. A visually-impaired person was asked to choose the meal from the menu of a catering service. The assistant of the visually-impaired person had the experience of transliteration from the printed books to the audio recordings. As the result of our analysis, the conversation held between two persons can be modeled with the same structure as the shopping cart systems on the popular web sites. The technique of transliteration was effectively used in the dialogs. The overall views and the interactivity were also important factors for reduction of both mental workload and required time.

A fast and intelligible text-to-speech application is required for visually impaired persons to acquire information effectively with screen reading software. In addition to engineering factors, we expect that experience have an important role for hearing. Ultra-fast speaking rate experiments of listening random 4 digits numbers with text-to-speech engine GalateaTalk synthesizer was carried out to observe learning effects.

The text-to-speech synthesizer that makes easy-to-listening voices at very fast speaking rates is expected to help visually impaired persons to acquire information effectively with screen reading softwares. We realized a fast and intelligible text-to-speech engine, using GalateaTalk synthesizer with the corpus with fast speaking rate. Our preliminary experiments of listening random 4 digits numbers showed that the statistical models trained with the fast speaking corpus was effective, and prominent learning effect was observed and the effect sustains for at least several days.

Communication between people is carried out fluently with voice. It is quite different from the current communication between people and computers. This fact shows that natural spoken language may contain rich information which plays important role in conversations. Non-verbal information, such as gazing and nodding, also plays important roles for natural and smooth conversation. From this point of view, we aim for the effective use of spoken dialog systems for visually-impaired persons. Our research deals with two tasks ; the use of web sites, and the use of home electric appliances, such as video recorder, electronic oven, air-conditioner, and so on. In this paper, we firstly give the outlines of these tasks. Then we investigate the characteristics of the visually-impaired persons, and their difficulties to perform these tasks. Finally we propose some possible solutions for these problems, and show some plans of fundamental researches that support the design of proposed interactive systems.

Visually impaired (blind) people use a computer with its text-to-speech (TTS) functions. Their needs for TTS functions are different from the ones of speech dialog. The authors have developed a new bilingual, Japanese and English TTS system based on Emacspeak to meet the requirements of Japanese visually impaired computer users. The current system, BEP (Bilingual Emacspeak Platform), can speak English with native English or Japanese English pronunciation and has three bilingual modes to speak bilingual information. The current paper firstly summarizes the requirements of Japanese visually impaired users for TTS functions and then describes the outline of BEP.

The Bilingual Emacspeak Platform(BEP) is a bilingual, Japanese and English, text-to-speech system for Japanese visually impaired Emacs users. Making full use of helpful functions of the existing Japanese screen-readers and trying to fulfill the requirements of advanced users, we are developing bilingual version of Raman's Emacspeak and bilingual speech servers for both Windows and Linux. In this paper, the authors firstly review the features of existing Japanese text-to-speech system and then, based on the experiences of BEP, show the requirements of a bilingual speech synthesis system for efficient and practical use.

The Bilingual Emacspeak is the self-speaking Emacs editor that controls a bilingual, Japanese and English, speech server. The aim of the current system is to enhance Japanese visually impaired graduate students, researchers, and programmers to use computers at high levels. The current system, which is implemented as an extension of Dr. Raman's Emacspeak, processes information by Lisp to provide it in the best way for the sense of hearing. The current system has richer expressive power than the conventional speech systems and can be used as an effective speech interface to sighted users as well as visually impaired ones. The potential of Emacs with a dedicated speech feedback will be shown by the Bilingual Emacspeak for Windows.

We are developing script-based software with which the visually disabled can exploit Windows 98 as much as the sighted do. The Script-based platform is an extension of Windows Scripting Host(WSH) with large format GUIs for users with low vision and Auditory User Interfaces(AUI) for users without vision. The Voiced Scripting Host system enables to use standard Windows 98 utilities and applications by using VBScript languages and create barrier-free programs. Through them visually disabled users can write programs to manuplate Windows Applications. We have added animations(Microsoft Agent) and making TV programs(TVML) to attract high school students and young potential users to Voiced Scripting Host system.

The Bilingual Emacspeak is the text-to-speech system that adds a bilingual, Japanese and English, speech output to an Emacs editor and runs on Windows and Linux. The aim of the current system is to enhance Japanese visually impaired graduate students and researchers to use computers at high levels. The current system, which is implemented as an extension of Dr. Raman's Emacspeak, can provide information in the best way for the sense of hearing. The current system can handle Japanese and English to meet the requirements at workplace and Internet. The first version of Windows porting of Bilingual Emacspeak was developed and some tests by a limited number of users have began. The structure, functions, and feature is discussed in this article.

The Barrier-free information processing system is a system that enables the disabled to access information regardless of their disability. Computers are useful for the visually disabled because computers work as their eyes by reading aloud E-mail and Web documents. We discuss the requirements for a barrier-free system in terms of visual disability. We are developing a new system, Voice Windows, that enables Japanese visually disabled university students and professionals to use personal computers at higher levels. Voice Meadow is a Meadow (multilingual Emacs) with a speech interface and Voiced Scripting Host offers basic components easily used by scripting languages. We also discuss the characteristics of the current system.The Barrier-free information processing system is a system that enables the disabled to access information regardless of their disability. Computers are useful for the visually disabled because computers work as their eyes by reading aloud E-mail and Web documents. We discuss the requirements for a barrier-free system in terms of visual disability. We are developing a new system, Voice Windows, that enables Japanese visually disabled university students and professionals to use personal computers at higher levels. Voice Meadow is a Meadow (multilingual Emacs) with a speech interface and Voiced Scripting Host offers basic components easily used by scripting languages. We also discuss the characteristics of the current system.

We are developing a voice-based platform that enables visually disabled university students and professionals to use personal computers at higher levels and write application programs on Windows 98. We find many softwares in market that assist the visually disabled in word-processing and browsing web pages. In university, however, students are required to advance to higher levels of computer use, including programming on standard OS's such as the Microsoft Windows and UNIX. Voice Windows is aimed to assist students and professionals in writing application softwares with standard audio user interfaces (AUI's) and graphic user interfaces (GUI's). Voice Windows focuses on use of script languages, such as Visual Basic Script (VBScript), Visual Basic Application (VBA), and Emacs-Lisp. They are interpreted line-by-line and allow access to most of Windows 98 standard utilities. Voice Windows supplies a voicified set of I/O interfaces, utilities, and help files, all with large-format GUI's for users with low vision. Through them, visually disabled users can write VBA codes to manupulate Microsoft's Word, Excel, and Access. The human interfaces are designed to be equally friendly to the sighted and the sight-disabled, and the programs developed on Voice Windows will be suitable for use in public terminals of community centers and libraries.