THE OTHER HALF OF THE EQUATION PC-BASED READING SYSTEMS A COMPARATIVE REVIEW by David Andrews From the Editor: David Andrews is the Director of the International Braille and Technology Center for the Blind at the National Center for the Blind. In the August, 1993, issue of the Braille Monitor he offered his evaluation of stand-alone reading machines. This is what Mr. Andrews now has to say about PC-based reading systems: In mid-1993 the International Braille and Technology Center for the Blind conducted comparative reviews of stand-alone reading machines--that is, computer-like devices that scan printed pages and turn their contents into synthesized speech. At that time we found that for simplicity of operation and ease of comprehension of the spoken word, the Reading Edge was probably the best buy. If, on the other hand, the user needed to read more complex documents with difficult layouts and a range of font sizes and print clarity, An Open Book was probably the best choice. This summary of our evaluation results is oversimplified and does not take into account the fact that updated software for a third machine, the Readman, is now available, which might change our conclusions if the assessment were repeated today. Now we have completed reviews of the three major PC-based (as opposed to stand-alone) reading systems sold today for use by blind people. We will review all three in this article: An Open Book Unbound from Arkenstone, Inc.; OsCaR from TeleSensory, Inc.; and the Reading AdvantEdge from Xerox Imaging Systems, Inc. One additional system has been shown at the past two NFB conventions, the Rapid Reader from Blindness and Visual Systems of Baltimore, Maryland. We ordered this system over a year and a half ago and still have not received delivery. It is impossible to review a system which you don't have and unwise to recommend a company with such unreliable business practices. What is a PC-Based Reading System? A PC-based reading system is similar to a stand-alone reading machine in that it scans the page (takes an electronic picture of it), analyzes that picture, and converts the text on the page into synthesized speech. Unlike a reading machine, a PC- based reading system, however, can do other things because it is based on an existing computer. All of the systems reviewed here require at a minimum a 386 computer with at least 8 megabytes of memory and adequate hard disk storage space to store the program and scanned documents. We recommend a minimum of twenty-five megabytes of free hard disk storage as a starting point. More free space is desirable since documents, image files, and swap files can get quite large. While a 386-based computer will do the job, a 486 will be better, and a Pentium-based system will be better yet because the reading process consists of three main steps. First, an electronic picture of a printed page is taken by the flatbed scanner. Second, this electronic image is transferred to the PC and analyzed, and any text is extracted. Finally, this text is converted into synthesized speech, refreshable Braille output, or large print output by the user's access device. The second step, the analysis of the image, is the difficult one--the one that relies on the power of the PC. So the faster the processor, the more quickly the analysis will take place. Using a Hewlett Packard Scanjet IIp flatbed scanner (the most common scanner around today), the scanning phase takes approximately fifteen seconds. A faster computer will not speed up this step. On the other hand, the page analysis can take anywhere from a second or two to several minutes. Factors affecting the length of time include the speed of the computer, the quality of the printed page, the amount of text on the page, the number of fonts employed, the size of the print, the presence or absence of graphics, the complexity of the layout, and more. Processing times on a 386 computer will seem painfully slow when compared to a fast 486 or a Pentium. We found that, when we used a ALR Pentium 90 with thirty-two megabytes of RAM, most pages had been processed and were being read before the HP scanner had returned the scanner head to the home position, and no page analysis took longer than about ten or fifteen seconds. If you are going to do a lot of reading, get the fastest computer you can afford. You will spend much less time waiting for the computer to process pages. With the use of the new Pentium computers, the scanning stage, not the analysis, becomes the time-consuming step. It used to be the other way around. For most people there are a number of advantages to using PC-based reading systems. If you already have an adequate computer and an adaptive access system, it saves money. If you wish to scan documents and save them as files for archival or editing purposes, then PC-based systems have an advantage over dedicated reading machines. PC-based systems allow you to use whatever access device or devices you already have--speech synthesizers, refreshable Braille displays, or large print display software and hardware. With most systems it is even possible to use multiple display methods simultaneously such as speech and refreshable Braille, or speech and large print. If you already have an adequate computer and an access device, it is clearly cheaper to purchase a PC-based system. Even if you don't, but you are prepared to shop wisely and take advantage of falling computer prices, you can still save money starting from scratch with a PC-based system rather than investing in most stand-alone reading machines. Moreover, you will have a fully functional computer that can be used for other purposes such as writing, doing your finances, or accessing the information superhighway. Finally, when manufacturers update their systems, PC-based reading systems are easier to upgrade than most stand-alone reading machines. Why A Special System? In the International Braille and Technology Center for the Blind, we are often asked, why do blind people need special reading systems like An Open Book Unbound or Reading AdvantEdge? Are they as good as the commercially available scanning products used in most offices? The answers to these two questions are related. Virtually all of the commercially available optical character recognition (OCR) software--the part of the package that extracts text from the page image--runs under Microsoft Windows. Until recently blind people have not been able to run Microsoft Windows. While there are now six Windows access programs on the market (with others due out in the near future, possibly even by the time you read this article), the programs have some rough edges and are still experiencing growing pains. Running an OCR system under one of them, while possible, would be difficult or impossible for the average user. The access companies (including Arkenstone, Xerox, and TeleSensory) have taken commercially available OCR systems and written their own control software that is speech-friendly. This is the software through which you initiate scanning, set and change options, read documents, etc. All of the companies use commercially available OCR engines. Arkenstone uses software from Calera Recognition Systems, which is also used in Calera's Wordscan Plus. The Reading AdvantEdge uses the same engine found in Xerox Imaging System's Text Bridge, and TeleSensory uses Omnipage Professional Version 5 from Caere Corporation in its OsCaR product. So systems for blind users offer the same OCR capabilities as their commercially available counterparts used by the sighted but are easier for us to use. What Do You Get? Arkenstone, Xerox, and TeleSensory wish to sell you an entire reading system, which includes a flatbed scanner, an interface card that goes inside your PC, and a cable to attach the two together, as well as all the software needed to control the scanner, perform OCR tasks, read documents, and set options. Each of the companies will sell you its software alone, which may save you a few dollars. If this is the route you decide to go, you must provide your own scanner, interface card, and cable; but these are all readily available in the commercial marketplace. Below we will describe each system, its layout, and operation. While not every feature will be mentioned or described, we will try to provide you with descriptions of the significant features and operations and to give you an overall feel for the operation of the software. Many of the features are similar, and all the systems scan fairly well, so some of the differences are subtle. An Open Book Unbound Arkenstone, Inc. Arkenstone calls its PC-based system "An Open Book Unbound" to differentiate it from their stand-alone product, which is called "An Open Book." The two products are virtually identical. An Open Book Unbound software-only package costs $995, while the software with a Hewlett Packard Scanjet IIIp scanner costs $1590. The Open Book Unbound system differs in one major way from its competitors. It takes direct control of your speech synthesizer or Braille Display and does not use your existing screen-review program or Braille-display control software. An Open Book Unbound actually runs under Microsoft Windows Version 3.1, although the user is never aware of this fact. The system provides him or her with speech-friendly or Braille-friendly prompts and output and takes care of all matters dealing with Windows itself. This is the reason, though, that conventional DOS-based access products are bypassed. Consequently, you must use a speech synthesizer or refreshable Braille display supported by Arkenstone. These include synthesizers in the Accent, Apollo, Double Talk, and Keynote Gold lines, as well as all of the Blazie products and the Audapter, DEC-Talk PC, Echo, Infovox, Sounding Board, and transport. The Artic SynPhonix line is not supported, although drivers have been promised since the beginning. Jim Fruchterman, President of Arkenstone, has assured us that Artic drivers are near completion. He says that they are waiting for software from Artic Technologies that supports indexing--a way for a synthesizer and its control software to communicate with each other about what each is doing. He further says that Arkenstone has hired a consultant to create its own software in the absence of software from Artic Technologies. Supported Braille displays include models from Alva, Baum, E.H.G., Frank Audiodata, Pappenmeier, TeleSensory, and Teiman. Fruchterman says that any Braille display that doesn't shut down when Windows is invoked should work. Further, any large-print display system that works with Windows should also work. Fruchterman says that they have made changes to support the Zoom Text product from AI Squared. An Open Book Unbound is controlled by using a series of menus. The functions of the system are normally controlled by using keys on the seventeen-key numeric keypad found on the right side of standard 101-key keyboards. Choices on Open Book menus are accessed by using four keys on the keypad, which represent arrow keys. The user moves up and down between items much as with many PC applications. Once an item is selected, its individual choices are accessed by using the right and left arrow keys. Choices are made by hitting what is called the Select Key, which is the five key--the one with the raised dot in the middle of the numeric keypad. The system has Beginner, Intermediate, and Advanced menus. The Beginner and Intermediate menus do not make all options available to the user and may be useful to the new user since there are fewer choices to confuse him or her. The sub-menus of the main menu include Read a Document; Library Options (where documents are stored, converted, and exported to floppy disk); Scanning Options (where contrast, page size, and the like are controlled); and Open Book Options (where you choose the menu level, set voice characteristics, exit the system, etc.). One drawback to this system, as it is currently designed, is that it takes a number of keystrokes to change some items. One must select an item, move to an individual choice, select that choice, and then press the Escape Key the proper number of times to back out of the menu system. Unlike An Open Book, An Open Book Unbound can also use regular keys on the alphanumeric keyboard. You can, for example, hit "O" for Open Book Options, "s" for Scanning Options, etc. You can also use the regular Escape Key for the Escape function and the regular Enter Key as the Select Key. The use of the regular keyboard is especially appreciated for typing document names, which is a tedious process at best on An Open Book Unbound. Once you know the system, these shortcut keys do help immensely. An Open Book Unbound has three basic modes: the Reading Mode, the Scanning Mode, and the Control Menus. When reading a document, one has to hit the Escape Key to leave the Document Reader prior to hitting the Scan Key to scan another page. Since the Scan Key is a dedicated key, it should be possible to go directly from Reading Mode to Scanning Mode at any time, but it is not. In fact, it should be possible to Scan at any time you are in the system. Jim Fruchterman, President of Arkenstone, assures us that this and other problems will be addressed in an upcoming Version 3. We were also assured by Arkenstone that some of the command awkwardness and oddities would be addressed in Version 2.0, which did not happen. In fact, changes have been promised since version 1.1. It is beginning to look as though changes will never be made and as though the longer the time and the more systems sold, the less likely are interface changes. An Open Book Unbound does have the ability to control contrast automatically. A Hewlett Packard Scanjet II or III series scanner is necessary to use this feature. Arkenstone was the first to offer this feature, and it generally works well and is very useful. You can also control the contrast manually for problem pages. There are a hundred levels of adjustment available when you choose a custom setting. There are also settings for Normal Contrast, Darken Page, and Lighten Page. The unit can also determine the direction in which the print goes on the page and read it, no matter what its orientation. This is a great time-saver for a blind person working alone. While it adds a few seconds to the recognition process, it is well worth the time to most of us, and the feature can be turned off. Different voices can be assigned to indicate normal, underlined, bold, and italic print. A different voice can also be assigned to the Menus. The Arkenstone software also has provisions for batch scanning--that is, one can scan (take pictures of) a number of pages all at once, then have An Open Book Unbound do the character recognition at a later time--like when you are at lunch. This feature can be valuable for big jobs, on slower computers, or to people with automatic sheet feeders. Once any text is scanned and recognized, it is automatically saved on the computer's hard disk. It is then possible to name a document and save it in a category. The software comes with a number of categories already set up, such as bills, books, brochures, general, letters, manuals, and recipes. It is possible to establish your own categories as well. The software should, however, be able to save or export files to a floppy disk. A number of steps are now required. In addition to batch scanning, An Open Book Unbound can scan and recognize a page, then read it or use the Quick-Speech Mode, in which speech is supposedly generated faster. However, it does not work very well. It is choppy and doesn't save much time. While the initial speech starts more quickly, the pauses and general choppiness use up any time savings. The software can save files in a variety of word-processor and other application formats--over fifty-five of them. One useful and unique feature of An Open Book Unbound is the ability to adjust the speed of reading on the fly, while it is reading a document, without losing your place. On the whole, installation is not a problem with this software, with two possible exceptions. The drivers for a number of speech synthesizers are on a supplemental disk, which you must put into the disk drive when prompted. You must then type in the name of the proper driver, including any serial port needed. If you do not know this information exactly, you can't proceed. Second, the installation of a flatbed scanner will be a problem to some. For more information see the "Strengths and Weaknesses" section at the end of this review. The software comes with an adequate manual in print, on tape, on disk, and in Braille upon request. The system also has a key-describer mode, which will be useful to new users, and the help system is good. OsCaR TeleSensory, Inc. TeleSensory, Inc., has just released Version 4.0A of its PC- based reading system, OsCaR. This version offers a number of improvements over version 3.0, including an improved menu structure, better optical character recognition, and some interesting new features, most notably the ability to correct individual words while reading a document. The OsCaR software costs $995, and the system costs $1,595 with a Hewlett Packard IIIp scanner. Supported scanners include the HP line; the GS Plus scanner from Datacopy, which is sold by Xerox Imaging Systems; and two Panasonic models. OsCaR, which is DOS-based, uses the facilities of your regular access system, regardless of whether it is a speech synthesizer and screen review program, a refreshable Braille display, or large-print software. However, the system does work best with one or more of TeleSensory's own products--Vert Plus, Vert Pro, Soft Vert, Screen Power, Power Braille, or the Navigator. We tested the system using Vert Plus and an eighty-cell Navigator refreshable Braille display, and the system tracked the menus flawlessly. In addition, the Braille display worked well and was always synchronized with the speech. We found it enjoyable to use the speech together with the Braille. The speech was good for long listening, and the Braille provided accuracy and precision in editing. The menu structure and operation of the OsCaR are more straightforward and easier to use than that of An Open Book Unbound. The up and down arrow keys move between main groups or among sub-menus and from item to item, while the right and left arrows show the options within each item. Once you move the right or left arrow to your choice, you don't have to hit Enter to select it or Escape to leave it. You can hit the down-arrow key to go to the next item. The previous choice remains in effect as it was left. When you get to the bottom of a sub-menu, the software beeps and wraps back to the top of that sub-menu. Further, when you first enter a sub-menu from the main menu, only the command letters are shown--that is, the first letter from each choice on that menu. If you know your choice, you can type that letter. However, if you are not sure, just hit the down arrow, and the letters will be expanded into words as you move past them. There are also function key choices to invoke some common operations from most points of the program, such as F4 to scan and F10 to quit. The first sub-menu on the main OsCaR menu allows the user to modify settings that affect accuracy. These include single or multiple columns; type of text (regular print, dot-matrix print, or faxes); page orientation; contrast; etc. Having all of these settings in one place is a good idea and a great improvement over Version 3.0. The next sub-menu is Advanced Scanner Settings. These include scanning mode, primary and secondary language choice, scanning from or to images on disk, page borders, etc. The secondary language option is an interesting innovation. It allows you to install a dictionary for a second language, like French or Spanish. If words in that language appear on a page with English words, the accuracy of their recognition should be improved by the presence of the dictionary. This feature should be a boon to those who handle bilingual materials. There are also primary dictionaries for U.S. and U.K. legal and medical terms, which should help people working in those fields. The Advanced Settings Menu is the place you can invoke batch scanning, in which you can scan images of a document or series of documents and do the recognition at a later time. OsCaR has an interesting feature for this process. If you scan with Batch Separators and put a blank sheet between the documents in a large batch, OsCaR will break the individual documents down into separate files. This feature would be useful to anyone reading a number of small documents--a student with a series of handouts, for example. Scan and Recognize is the next choice on the main menu. When OsCar is set up to scan and read a page, it does one thing differently from the other reviewed systems. It scans and recognizes the top of the first page, showing you the first few lines, and presents you with a sub-menu of choices which include reading the entire page, continuing to scan additional pages, and saving the current page. In some instances this is a useful feature because it allows you to see quickly whether you are getting useable text or garbage. However, there seems to be no way to defeat this feature, and with those documents in which you have no doubt about the useability of the scan, it can be annoying. Next on the Main Menu is Convert. This choice allows you to convert a scanned document to a specified file format. A full complement of programs and versions of programs are supported with over eighty-five choices. There is support for all versions of WordPerfect, including those for the Mac and Microsoft Windows; Microsoft Word (all versions); Lotus 1-2-3; Dbase; ASCII; and much more. Output can even be sent directly to a Braille translation program. Duxbury, Hot Dots, and Mega Dots are supported. The Files Menu is next. OsCaR allows you to do file management from within the program--that is, to change directories, copy and delete files, view ASCII files, save or print files, and more. You can also run an external program from this sub-menu, such as a word processor. Of course, you can also specify what file will be loaded into the program as it runs. These features work well and will be useful to some people. The next sub-menu retrieves and saves settings. The settings file maintains the choices made in all the menu options we have been discussing. It is possible to have separate settings files for different kinds of work. The next choice on the OsCaR Main Menu displays OsCaR settings. This gives you information about what the system is currently doing and about the page being scanned. The next-to-last sub-menu is Other. Here you can halt scanning, erase the current page, or manage your custom dictionary. If you encounter a word or phrase regularly that the system has trouble dealing with, you can enter it in the Custom Dictionary to improve the speed and accuracy of recognition. The final choice on the Main Menu is Quit OsCar. OsCaR comes with a good manual (available in print or Braille or on disk or tape). The help system is context-sensitive and quite good. The installation process was also easy to understand and went without problems. The software is copy- protected. It allows two installations. There is also an uninstall process so you can move the software from one computer to another, perhaps new, computer. Reading AdvantEdge Xerox Imaging Systems The Reading AdvantEdge (pronounced advantage) is the PC- based version of the Reading Edge, the company's successful stand-alone reading machine. It uses the same OCR core technology. However, unlike the Reading Edge, which strives to be simple, the Reading AdvantEdge packs a lot of power and many options into a conventional menu system. The manual for the Reading AdvantEdge is detailed and straightforward. It will provide too much detail for an experienced user but will be helpful to a new user. It comes in print and Braille and on cassette and disk. The installation process for the Reading AdvantEdge was easy to follow and went well. We initially had some difficulty installing a GS Plus scanner, which ultimately proved to be defective. The company replaced it quickly and without questions. The Reading AdvantEdge will work with any access system, since it is a DOS-based program. It supports two scanners that Xerox Imaging Systems sells, the GS Plus and the SA4-3. The GS Plus was originally manufactured by Datacopy, a company Xerox acquired several years ago. Though the scanner has now been discontinued, Xerox still has a number on hand and sells them at a relatively good price. The other scanner Xerox sells is the SA4-3. This is the large bookedge scanner that was used with the old Kurzweil Personal Reader. The Reading AdvantEdge also supports the Scanjet line of scanners from Hewlett Packard. The software-only version of the Reading AdvantEdge is priced at $795; the price with a GS Plus scanner is $1,295; and with a SA4-3 scanner the system will set you back a whopping $2,995. Unless you need a bookedge scanner, get an HP scanner on your own. The GS Plus scanner is slow, and with the drop in price of the HP IIIp scanners, it is no longer such a good deal. The list price for the HP scanner is $595, and a good shopper could find it for $500 or less. The Reading AdvantEdge uses a conventional menuing system with a Main Menu and sub-menus subordinate to it. In some instances sub-menus go three or four layers deep. Menus are navigated using up and down arrow keys or Tab and Shift-Tab keys. With some screen review programs you get double speech--that is, lines repeating themselves when the arrows are used. The use of the Tab and Shift-Tab keys as an alternative prevents this and is a convenience. There are also alternative keystrokes in other places to avoid speech problems. Once you reach an item you wish to change, the Enter key displays possible choices. Because of the number of sub-menus and, in some instances, sub-sub-menus, it would take several pages to describe the Reading AdvantEdge and all of its choices completely. Suffice it to say, it is a powerful and complete text and imaging scanning system with a full set of features and options. The software offers the user many more options if needed or desired. However, you do not have to delve into this complexity unless you wish. The Main Menu has seven choices: Scan, Batch Facilities, Utilities, Applications, Options, Help, and Quit. The first choice is Scan. When you press "S" or hit "Enter" to choose it, the system says "scanning Idle Menu" and starts scanning. We find this a little misleading. Reading AdvantEdge says "Scanning Idle Menu" because you can do a few things while there--abort the scan and enable or disable the Speak-Immediately utility. However, the primary purpose is to scan, and we were initially confused by the message, particularly in light of the fact that there is a long pause before the GS Plus scanner begins. We would prefer a "Scanning, Please Wait . . . " or similar message. As mentioned above, you can have the system speak immediately or go back later and read the scanned document. The manual says that the reading may not be quite as accurate with the Speak-Immediately utility because the decolumnization and other page formatting steps may not be as accurate. We believe that this is also the case with the other reviewed systems. Once in the utility, you can move around using the arrow keys or Alt- Key combinations. The utility also has the ability to search in either direction for text strings. Next on the Main Menu is Batch Facilities. Like the other reviewed systems, Reading AdvantEdge has the ability to scan (record images of) the pages and process them separately. These procedures are set up and initiated from this Main Menu choice. The next choice is Utilities. This is where you save and retrieve settings, read ASCII files, and do file management tasks such as copying and deleting files. Applications is the next choice on the Main Menu. the first application is the Text Browser, a program which reads ASCII files. There are also four blanks for plugging in the names of applications you wish to run from within the Reading AdvantEdge. These could be a word processor, Braille translation program, etc. You can also automatically or manually specify a filename at the time the program is loaded. You can invoke any of the applications using Function Keys from anywhere within AdvantEdge. Options is next on the main menu. This area is probably the most complex area of the system since some of the choices go down several layers. There are sub-menus for Text Scanning Options, Formatting Options, and Image Scanning Options. Under Text Scanning Options, for example, there are sub-menus for Recognition Options and Formatting Options. Under Recognition Options you can set document type, brightness, text type, user lexicon, language, questionable character threshold, eliminate halftones, and maximum point size. This list of choices illustrates the power of the system. The User Lexicon is a user- defined dictionary for acronyms or other words that the system might have problems with. The text type can be set to Alpha- numeric, that is letters and numbers; Numbers Only; User Lexicon Only; or User Lexicon with Numbers. For example, let us say that you work in a warehouse and have to scan inventory sheets. You could define the items you stock in a User Lexicon and set the Text Type and Filter options so that only numbers and your parts showed up in the scan. As you can see, there is lots of power and sophistication in the software. Under the Page-Control Options, which is under Text-Scanning Options, you can set Page Orientation, Document Feeder Presence, Columns, Cropping Window, Units of Measure, and Width and Height of the page. Some of these choices can also be set in other parts of the software, such as in Batch Facilities, adding further to the number of possible choices. Under Formatting Options, which you recall is one of the three main choices on the Options Menu, you can pick one of forty-six possible file conversions. There is a good selection of applications and file formats available. You can also control how formatting information in the source document is handled in the output. There are choices for centering, spaces and tabs, end of line and end of page markup and many other things. The creative hacker could tinker and perform some pretty specialized formatting with these options. It would, for example, probably be possible automatically to substitute the codes used by a Braille translation software program for formatting information found in a document. The final choice on the Options Menu is Image Scanning. The software handles TIFF and PCX files, and a full set of options is available. The last two choices on the Main Menu are Help and Quit. The help system is context-sensitive and complete. Scanning Results Now let's get to what most people consider the important part of our testing: how accurately these software packages scan and read documents. We have developed a set of standard documents that we use to test optical character recognition products. Those used here are the same as the ones we used in our previous tests of stand-alone reading machines. These include a poor-quality, dot-matrix printout--one produced with a 9-pin printer and bad ribbon; a letter-quality printout; a laser-printed sheet with different sized serif and sans serif fonts, ranging from six to fourteen-points in size; a fax; a flyer from Egghead Software; and a page from a bank statement. The flyer has three columns, graphics, multiple fonts, and different colors and represents a fairly complex page layout. We have added some additional pages, which include letter-quality text skewed at an angle of five, ten, fifteen, and twenty degrees to test how well the systems read crooked documents. While these tests are not exhaustive (they do not test how well the systems preserve formatting information, for example,) they are practical. The test pages represent a cross section of the kinds of documents each of us wants to read. For these tests we have also instituted a scoring system. Scanning most printed pages produces text that falls into one of five possible categories: 1. Garbage: random characters and indecipherable results. 2. Poor text: mostly garbage with a few understandable words sprinkled randomly throughout. 3. Fair text: mostly understandable text, but still a relatively high number of mistakes. Familiar or simple text can be followed, but unfamiliar or difficult material or material with numbers or computer commands may not be useable. 4. Good text: very understandable results with only a few mistakes; problems may occur only with numbers or computer commands and the like. 5. Excellent text: few if any mistakes, no more than one or two per page. We scanned each page and assigned the appropriate number to the results. For the typefaces page, which has ten print style- size combinations, and the skewed pages, which have four results, a result number was assigned to each combination or page; and the resulting scores were added together and divided by the number of items in the group (ten and four respectively) to determine the composite score. We scanned each of these pages with each machine. We started with the default settings and changed them only if the results were bad or if there was a specific setting for the type of page being scanned (fax or dot-matrix, for example). So, unless otherwise indicated, brightness, contrast, and print style settings are at their default or automatic values for all tests. Finally, we scanned all pages with the Reading AdvantEdge three times. Like previous systems from the company, the software does better once it has trained itself on a page. Multiple scans seemed to have no effect with An Open Book Unbound and OsCaR. To end the suspense, here are the scores. Then we will try to explain the results. The winner, with 30.8 points, is Arkenstone's Open Book Unbound. The second place system, with 26.8 points, is OsCaR from TeleSensory; and Reading AdvantEdge, with a score of 22.8, came in third. While these differences may seem significant--and there certainly were some notable differences--we stress that our results should not be regarded as the last word. Depending on the kinds of documents you regularly read, these results may be meaningless to you. To be certain, you should conduct your own test on all three machines, using a set of documents you normally read. Below is a table listing the results for each machine and test: SCANNING TEST RESULTS Open Book OsCar Reading Advantage Letter-quality 5 5 5 Dot-matrix 5 5 4 Fax 4 4 4 Flyer 4 4 3 Typefaces 4.8 4.2 3.8 Skewed pages 4 2 1 Bank statement 4 2 2 Totals 30.8 26.2 22.8 As you can see, all of the systems read the letter-quality page perfectly. They also read the dot-matrix quite well, although the Reading AdvantEdge stumbled a little, even though we changed it to its dot-matrix setting and ran the page through three times. All of the packages did pretty well with the fax, stumbling a little on the letterhead and at the bottom. Each received a score of 4. The Open Book Unbound and OsCaR received a score of 4 on the Egghead Software flyer, and the Reading AdvantEdge got a 3. They all had trouble at the top, but in general it was possible to follow the sense. Overall, the Reading AdvantEdge missed more words, which resulted in its lower score. The type faces page had two columns, one of Roman type, one of San serif type, ranging in size from six to fourteen points. In general Open Book Unbound, which received a 4.8, read the smaller sizes better and had no problems with decolumnization. The OsCaR system got a 4.2 and had a little decolumnization trouble. It didn't do as well with the smaller sizes. Reading AdvantEdge, which received a 3.8, had some decolumnization problems along with problems in the smaller print sizes, and it also stumbled some right up through the Roman ten-point size, which the other systems were able to handle. With its Version 2.1, Arkenstone says that it can handle skewed pages up to fifteen degrees. We could only get it to handle a ten-degree skew, which resulted in a score of 4. OsCar was able to handle a five-degree skew, netting it a score of 2, while Reading AdvantEdge couldn't read any of the skewed pages, resulting in a score of one. Finally, there was the infamous bank statement. As you may recall from the stand-alone tests, the results were disappointing. This time An Open Book Unbound scored a 4. While we would still not recommend using the results to reconcile your bank account, it did read the majority of the information accurately and did a considerably better job than the other two systems, which each received a score of 2. Strengths and Weaknesses As we pointed out earlier, our scanning tests should not be regarded as conclusive; you should certainly conduct your own if possible. Each of the systems has its own strengths and weaknesses. Further, there are other considerations, such as price, special discounts, the reliability of your local dealer, the equipment your friends are using, the access equipment you have, your experience with systems at work, and more. An Open Book Unbound from Arkenstone is relatively easy to use. It is easy to set up and get into service quickly. This reflects the stand-alone-reading-machine origins of the product. The system did win our scanning tests, and in day-to-day use and demonstrations with a variety of materials brought in by people on tour, we have found the system to be slightly more accurate overall. On the negative side, the menu system seems somewhat cumbersome to use when making changes to the system, and your speech synthesizer may not be supported, although most are. Also it is not very easy to save and export files, and some may find the Library system of pre-defined categories awkward or restrictive. The OsCaR system from TeleSensory is particularly strong when it is used with the company's own access systems. Everything speaks well and is exactly synchronized. Overall, the system has good accuracy, and the revised menu structure is an improvement over version 3.0. The word correction feature can be quite useful, as can the user dictionary and second language. On the negative side, TeleSensory is a company in flux. The company has changed management twice in less than two years, and rumors persist that the blindness-products division will be spun off or sold. Some people have reported support and service problems with TeleSensory, though we have not had any such problems. If you ask around enough, you will hear good and bad stories about any company, particularly the larger ones. The representative in the Baltimore area has been willing and helpful, although this varies from place to place. Finally, it may take a little work to get OsCaR or Reading AdvantEdge to behave with your speech system, although on the whole they do well. Reading AdvantEdge is a sophisticated and powerful system. There are dozens of settings that can be user-adjusted if you wish, but they don't have to be. If you wish to tinker, this is the system for you. While Reading AdvantEdge didn't do as well in our scanning tests, it read basic materials well and would probably do a good job for most users. Greg Guidice, marketing manager for the product, reports that Xerox is continuously working on optical character recognition improvements, which will be offered as upgrades as they become available. On the down side, the number of menus and choices may not appeal to everyone. We found it easy to get lost at times. It can be difficult to find one obscure setting if you are not sure where you are. Stephen Baum, the author of the software, has made a real effort to ascertain what blind people want. He is, or at least was, active on NFB NET, CompuServe, and other BBS's and tries to respond to questions and input. He has also written some useful utilities which have been released to the public domain, including a program to identify money and a utility to strip headers and footers from text files. Baum has left the company, although he says that he still might be working with Xerox on a consulting basis. One possible drawback with all the systems is the need to install a scanner and its interface card. This can be tricky because the interface is achieved by what is called a SCSI card, (pronounced scuzzy) which stands for Small Computer Systems Interface. SCSI adapters require their own memory addresses, interrupts, and the like. Further, the memory space they use may have to be excluded from the addresses controlled by a memory manager such as EMM386 or QEMM. If this all sounds like Greek, you will probably need some help installing your scanner. The Open Book Unbound manual provides little assistance in this area although there are some examples, and many dealers will do installation for you. The OsCaR and Reading AdvantEdge manuals do provide some assistance, particularly the Reading AdvantEdge manual for the scanners the company sells. In fact, there is a separate installation manual, which is available in Braille. As mentioned above, some dealers will help you with scanner installation. If you don't know something about installing expansion boards in your computer and about memory management, you really should plan to seek assistance from a friend, co- worker, or dealer. Conclusions We are asked almost every day, "What should I buy?" Well, it depends. As you can see from the above reviews, there are many factors to be considered. There is no one system right for everyone. That in fact is part of the reason that the National Federation of the Blind created the International Braille and Technology Center for the Blind. We have and use all of the systems under one roof so that we can work with them and get an idea of their individual strengths and weaknesses. We can then write reviews like this one and help you decide what is the best system for you. There has been more change in the scanning and OCR area than in any other arena in the access technology field. Prices have steadily dropped, both for stand-alone machines and for PC-based systems, and the technology has improved dramatically. Overall, all of the scanning results were somewhat better than those we obtained a year and a half ago when testing stand-alone reading machines with the same documents. The OCR field is in flux at present. In late November an announcement was made that Caere Corp., the company that provides TeleSensory's OCR engine, had bought Calera Recognition Systems, the company that provides Arkenstone's OCR engine. Jim Fruchterman, President of Arkenstone, admitted that they are certain that they will get Calera's next upgrade, which will be used in Arkenstone's Version 3.0, but after that they just don't know. No announcement has been made about the future of Calera's products and whether or not they will be maintained as separate lines. What this merger means to us as blind consumers is impossible to determine at this point. For Further Information For further information or to find out who your local dealer is, you can contact one of the following: Arkenstone, Inc., 1390 Borregas Ave., Sunnyvale, California 94089; (800) 444-4443 or (408) 752-2200. TeleSensory, Inc., 455 North Bernardo Ave., P.O. Box 7455, Mountain View, California 94039-7455; (800) 227-8418 or (415) 960-0920. Xerox Imaging Systems, Inc., 9 Centennial Park Drive, Peabody, Massachusetts 01960; (800) 343-0311 or (508) 977-2000. For further information or data on updates and recent developments, you can also contact the International Braille and Technology Center for the Blind. You can reach us at the International Braille and Technology Center for the Blind, National Federation of the Blind, 1800 Johnson Street, Baltimore, Maryland 21230; (410) 659-9314. You can also reach us through NFB NET, our computer bulletin board system, by calling (410) 752- 5011. You can also reach David Andrews via Internet at da0011@epfl2.epflbalto.org. Reprints of the stand-alone reading machine reviews can still be ordered in Braille or print from the NFB's Materials Center or the entire Braille Monitor for August, 1993, can also be ordered for $2, in print or Braille or on cassette or talking- book disk. The Braille Monitor and the reprint can also be downloaded from NFB NET.