2000 Conference Proceedings

Go to previous article 
Go to next article 
Return to 2000 Table of Contents


Dueling Scanners 2000

Dueling Scanners 2000 was held at the Technology for Persons with Disabilities (CSUN) conference on Thursday, March 23rd. Although the session was held at 8 AM, attendance and audience participation were high. Before beginning the actual report, some things should be explained to those who did not attend this session. First, the session was not as long as it has been in previous years. This years' session lasted a bit over two hours, rather than the three hour sessions that occurred in previous years. Second, it should be noted that the information contained in this report is limited to the framework provided by the six questions as agreed upon by both of the participating vendors and the judges.

This report receives wide circulation, and accordingly, it is important that we present as much information as possible. However, there is no substitute for hands-on experience with technology, and this report should only be viewed as a source of information, and not the only factor in making a decision. We believe that this report may assist rehabilitation professionals, educators, and their clients, however all of these individuals and anyone else in the blindness field should not only consider the data presented herein but their own experiences and the needs of their clients.

The Vendors

The two leading vendors of Optical Character recognition software for blind and visually impaired users were represented by the following individuals:
Arkenstone Inc.
James Fruchterman, President
David Offen
Beth Thomlinson
Kurzweil Educational Group
David Bradburn
Stephen Baum
The Judges
Larry Skutchan, Director of Technology, American Printing house for the Blind Louisville Kentucky.

Richard Ring, Supervisor, International Braille and Technology Center for the Blind, National Federation of the blind Baltimore, Maryland. As in the past, the session was ably moderated by Dr. Peter M. Scialli of ShrinkWrap Computer Products.

The systems used in this years Dueling Scanners were as follows: Arkenstone brought a Pentium II 450 with 64 MB ram, and n HP5200C scanner. Kurzweil brought a Del Pentium ii 450 with 64 MB ram, and an Epson Perfection 1200 scanner. It should be noted that since the Epson Perfection 1200 scanner is much faster than the HP5200C, timing issues would have been somewhat skewed were they to be considered in this report. Realistically, when a computer that has a processor speed of 450 MHZ or better is involved, timing issues are not a significant factor.

The software used was, Openbook Ruby version 4.02 and Kurzweil 1000 version 5.0. Arkenstone’s entry is an improved release of the Ruby 4.0 software shown last year. Kurzweil’s entry was released as a major new version the day prior to Dueling Scanners.

The session began with a discussion of those features that allow the respective software packages to be used by novice computer users. This discussion raises an issue which we feel is one of the most critical when speaking of optical character recognition/reading systems. The significance of this question cannot be over-emphasized, since for many of the individuals who are using this kind of software, its operation will be their first exposure to computers.

Installation

Kesi described the ease of installation that they felt their product possessed. We agree that the installation process is fairly straightforward, and that it could easily be accomplished by a novice user. In fact, both K1000 and Openbook Ruby are quite simple to install, and if the user wishes to do so, he/she can accept the default installation offered by each program. Kurzweil1000 will, if no user intervention is forthcoming, install itself. Kurzweil also pointed out that if desired, K1000 could be installed so that it would start when the computer was booted up. This would be useful only if the computer would not be required for any tasks other than scanning and reading. Both programs provide the user with software-based text-to-speech engines that require a sound card. Kurzweil1000 comes with both Flextalk and RealSpeak. Openbook is furnished with IBM’s Via Boice Out Loud software.. Both products support the use of hardware synthesizers by taking advantage of the "SSIL" (speech synthesizer interface library) standard which was developed by Arkenstone. During its installation, Kurzweil will automatically detect the presence of any Sapi compliant text-to-speech engines one may have on a computer system, and therefore the user can easily switch from one synthesizer to another. Openbook, during its setup, permits the end-user to customize the program so that synthesizers may be selected from a list of several dozen currently available speech synthesizers.. The advantage here is that the product need not be reinstalled if a new voice synthesizer is employed sometime after Ruby’s initial installation.

Once the respective software packages are installed, what provisions a re made to accommodate the first time computer user? Kurzweil addressed this issue first, and many of their points are well taken. They stated that one of the challenges facing a novice user is becoming accustomed to hearing documents read with synthesized speech. This is a serious issue, it has been our experience that computer systems are often found wanting by beginning users based solely on the quality of the speech. Kurzweil pointed out that with the introduction of "RealSpeak" in version 5.0 of their software, they believed that they have provided a far more listener friendly experience for the novice user. It is difficult to comment here, because, though "RealSpeak" has many qualities that make it sound somewhat more human than any other software speech synthesizers, it is extremely difficult to know what someone else's listening experience will be like. The "RealSpeak" voice is based on actual recorded segments of human speech rather than the traditional method of generating synthesized speech which uses the sounds of vowels and consonants (phonemes)in order to produce speech. The fact of the matter is, the only way in which one can determine just how a new user would adapt to a specific software speech system is to allow them to hear it and become familiar with its particular cadences and rhythms

Kurzweil 1000 has made it easy to use the manual as a tool for instruction. When the user brings up the on-line manual, he/she can access a list of bookmarks, which, to the user, is just as if he/she were reading the "table of contents". All one then has to do is to use the arrow keys to find the particular topic of interest, and press "enter". It is true that the majority of users will probably not read very much of the manual, but it is furnished, we believe, in a format that allows for quick reference.

Openbook Ruby and Kurzweill1000 provide, upon request, Braille versions of their documentation. Openbook provides their manual on cassette, on the software cd, and on line. However, with Openbook Ruby, it is not as easy to quickly locate the specific topic one wishes to obtain information about when reading the full text manual. There is no list of specific topics that a user can scroll through. However, there is a quick command reference provided from within the program and in a braille document included with the software package. Kesi does not furnish their manual on cassette, and we believe this is something that they should consider, because, unfortunately there are far too many blind and visually impaired persons who do not know Braille.

Kesi also furnishes a 17 key key pad, and the product can be used with a "game" controller as well. If a "game controller" pad is used, there will be fewer keys, and, correspondingly, fewer functions available to the user. Another feature that we feel is beneficial is the "help and status" function. By pressing the "help and status" key, and then pressing the key one wants help on, the user can obtain either a short or long help message pertaining to the key that has been pressed. This is, in one sense, a context sensitive mini-tutorial to learn the product as you go, which when it comes to computers is probably the best way to become familiar with software and its functions.

Openbook Ruby also provides several methods for obtaining context-sensitive help that will allow the user to learn not only the keys required to perform routine tasks, but also a way to quickly gain access to further information. When one presses the "help" key, the "dash" key on the numeric keypad, a message will be spoken telling the user something about the movement keys and other important keys. After hearing this message, a user can choose to either close the help system, or to explore further help topics.

Openbook also provides a "key describer" mode which, when active, will allow a user to press any key and hear a brief message describing its function.

Two important factors that distinguish K1000's appeal to the novice user are: the products "auto name" feature, and its "auto load" feature. A novice computer user tends to have no concept of files, folders and subfolders. Even the idea of naming files (or, for that matter, the need to do so) is beyond many who have just begun to use computers.

The Kurzweil1000 will automatically give each scanned document a name, usually consisting of its first line. This means that the beginner need not be concerned with the naming of files, and though a user might well decide fairly quickly that the ability to name files is a useful thing to know, he/she need not be concerned over much about this until he/she becomes more comfortable with computer use. Arkenstone addresses the file naming issue in an entirely different manner. With Openbook Ruby, the user need never name a file, so long as he/she is aware of the fact that unless some file name is chosen, all material scanned will be saved in a file called "untitled.ark", which will, unless the user learns to name files, be the only file that contains the scanned documents. Although, in a sense, it accomplishes the same goal, e.g. eliminating the need for the user to understand file and folder naming conventions, we believe that Kurzweil 1000 has come up with a better way to deal with this issue. Another feature that is available to a novice Kurzweil1000 user is "autoload". When "Autoload is enabled, all of the files that were opened during a given Kurzweil1000 session will be opened again during the next session. This means that a user who is not aware of how to name and open files will not have to do so. Arkenstone’s approach is to make quickly available the five most recently read or created documents, some or all of which may be selected from a File menu. For the novice user who is interested in only scanning and reading, Arkenstone discussed a solution that might well be worth consideration: Arkenstone sells a software version of "Vera" (Very Easy Reading appliance) which is the same software that is running on Arkenstone's stand-alone reading machines. The next question discussed was how both products deal with the issue of file formatting and saving. Both products provide the ability to set up launchable applications, that is, programs that can be started from within Openbook Ruby or Kurzweil1000. Kurzweil1000 allows the user to set up such choices at any time from within the program itself, while with Openbook Ruby, this is done during installation. If, at any time, a user of Openbook Ruby wishes to alter the list of launchable applications, he/she must insert the Openbook Ruby CD and use a setup program to add them.

Open Book Ruby and Kurzweil1000 both allow the user to specify that an open document be passed to certain launchable applications in a specific file format automatically, e.g. Microsoft Word. Both programs allow the user to save files in a number of file formats quite easily from the "save as" item on the file menu. Kurzweil1000 has 140 possible file formats, but only a short list of these formats will be shown unless the user specifies that he/she will require a more detailed list. Most users will never need all of the file formats that Kurzweil1000 provides, but for those who do, they are there. Both products will save files in the most common formats such as Microsoft Word, Corel Word Perfect and ASCII text to name but a few. The process of saving files in a format that can be used by other applications is, we believe, straight-forward in both Openbook Ruby and Kurzweil1000, We believe that the ability to not only launch a specific application but to pass an open document to that application in the proper format is an excellent feature and we are pleased to see it in both products. It should also be noted that one of Kurzweil1000's new features is the ability to automatically back-translate a Grade II Braille file. This means that if a file is imported into the Kurzweil1000 with an extension of "brl", "brf" or "bfm", Kurzweil1000 will recognize such a file as a grade II Braille file, and back-translate the file into ordinary text. Similarly, when one opens a PDF file with K1000, the system extracts the text from the document and makes reading it very easy.

The Vendor Documents

All that has been discussed thus far in this report is worth considering, but for those of us who use Optical Character Recognition on a daily basis, the most important issues involve the ability of each of these packages to accurately recognize text. The vendors of both products brought documents that they believed would best highlight the strengths of each package. Arkenstone brought a magazine article which featured a mixture of text and graphics. This is a good test of any OCR package, because numerous magazine articles will present such pages. In this instance, the document that Arkenstone used in its test featured text that was part of a photograph of a well-known political figure. This particular piece of text was displayed on the political figures thigh. Openbook Ruby was able to read this page in a sensible and logical order. Kurzweil did not do quite as well, it was unable to discern the text that was part of the photograph. Kurzweil Educational Group brought a document which would highlight one of Kurzweil1000's new features, the ability to read tables one cell at a time. This is an exciting new feature, and when the document was scanned, Kurzweil1000 did an admirable job reading the table. The table presented computer & electronic information, and Kurzweil1000 spoke headings and cell data logically and efficiently. This resulted in the first round of applause heard during the session. When Arkenstone's turn came to scan this document however, we were quite surprised by the results. Though Arkenstone's Openbook Ruby edition does not formally support table recognition, Ruby did amazingly well when it came to reading this document. We must emphasize here that Ruby did not render the document as well as Kurzweil1000, but Ruby presented the tabular data in a logical and understandable manner.

Documents furnished by Judges

During this year's session, because of time constraints, only one document furnished by the judges was scanned. This document was an advertisement for an electronic door system called, the "Out Door". It presented a challenge, in that it consisted of several fonts, and its data was presented in a tabular format. Again, the results were surprising. Both products rendered a fairly readable version of the document, but it must be said that Openbook Ruby did an extremely good job when we consider that it doesn't currently support the explicit reading of tables. In fact, Kurzweil’s table reading mode was slightly fooled into including some surrounding text as part of a table heading. It should be pointed out, however, that reading structures like tables is a somewhat subjective task, and even some humans can be fooled by some table formats.

Low Vision Considerations

Both of these products incorporate features designed to accommodate users with low vision. Though both contain a rich set of features, we believe that Openbook Ruby provides a slightly better set of functions to make the use of their software easier for those with low vision. Openbook Ruby includes such features as "Exact View" which preserves the page just as it was scanned. This feature, allows the low vision user to see not only the text from a scanned document, but any graphic present on the page as well. Both the text and graphics may be magnified in Exact View mode. Openbook Ruby also allows the user to determine how much text is displayed on a screen. One can have as little as one line of text, or three quarters of a screen. In conjunction with this feature, Ruby allows the user to not only increase the font size, adjust foreground and background colors, but to increase and decrease the spacing between such elements as characters, words, and sentences. The preferred font as well as its size can also be changed. Kurzweil 1000 also gives the user the ability to magnify text on the screen, highlight the text as it is spoken, and it allows the user to change the default font as well as its size. The highlight color and the block color can also be changed. We should add here that both of the judges are totally blind, and we base our opinion of the low vision features upon the number of features and the manner in which the features can be combined and varied.

Braille Production

Both programs provide a way to scan a document and produce grade II Braille. Both products permit the entire process to be accomplished quickly with just a few keystrokes. Kurzweil1000 incorporates a Braille Translation program derived from NFBtrans, a public domain Braille translation package originally developed by the National Federation of the Blind. Once your Braille embosser has been set up and is functioning under Windows, Kurzzweil1000 can translate and emboss scanned documents. Of course, this method does not yield perfectly formatted Braille, but if quick and dirty Braille will suffice, this method streamlines the process of taking scanned documents and translating them into Braille. Openbook Ruby, on the other hand, is supplied with "Turbo Braille" another freeware Braille translation program, however, it is not automatically installed along with Openbook and therefore it must be installed manually. The manual installation of Turbo Braille does permit some ability to tailor the braille output to specific embosser models. For ideally formatted braille, however, it is probably best to consider the purchase of a full featured braille translator as an addition to either reading system. Openbook Ruby will, during its installation process, detect the presence of the major Braille translation products such as "Duxbury" and "Megadots" and will include these products in its list of launchable applications. In contrast to Kurzweil’s approach, to translate a document into Braille with Ruby, a Braille translation program must be launched and the document in question translated and embossed. Braille translation programs such as the included Turbo Braille or the commercially available Megadots make it possible to automate the process such that simply launching one of these from Ruby can result in immediate translation and embossing. K1000 will, of course, permit third party braille translators to be launched as well.

Multi-Language Support

Both Kurzweil1000 and Openbook support numerous recognition languages. However, as demonstrated at Dueling Scanners 1999, Kurzweil1000 is unique in its ability to automatically detect languages and load the appropriate text-to-speech engine and read the language detected. Openbook can currently recognize fourteen languages, and it is furnished with a number of text-to-speech engines (the six speech synthesizer languages included are part of the ViaVoice Out loud set of languages Openbook cannot automatically detect the language being scanned, nor can it switch from one text-to-speech engine to another without human intervention. Neither product will automatically install more than one speech engine, if other languages are required, the respective text-to-speech engines must be installed manually.

The Future

The final question dealt with what the respective vendors felt was possible in the future for their products. Kurzweil spoke of the possibility of using digital cameras in order to scan documents, improved speech recognition, and the possibility of instantaneous translation from one language to another. Arkenstone talked of a time when the Internet and most software applications will be completely integrated. They discussed the availability of more free tools, e.g. freeware and software, as well as the ability to interface software with devices that have been designed with accessibility in mind.

Conclusions

It is a tremendous opportunity to attend and judge the Dueling Scanner sessions. Those who attend and even the vendors can learn from the sessions, and a good deal of information is presented. However, when the smoke has cleared, with what are we left? Both of these products are well thought out and well designed. They both provide an excellent solution for professionals and home users alike when it comes to Optical Character recognition. However, how does one make an informed decision as to which of these two products will suit the needs of the individuals who will require them? Let us look carefully at some of the features provided by both products. It is beyond the scope of this report to list all of the features contained in each of these programs, but if features are your primary concern, it is clear that Kurzweil1000 has a greater array of features designed to enhance the overall scanning and reading experience. We have mentioned many of these features, however we should like to point out that we have only scratched the surface. For example, with Kurzweil1000, one can use its "document summary" function to create a summary of any scanned document. Kurzweil1000 also provides voice recognition, meaning, that some of the commands that one would most frequently use within the product can be spoken and understood by K1000 provided a microphone is properly installed. Kurzweil is also furnished with two complete optical character recognition engines. The first of these is the RTK developed by Expervision, and "Fine Engine" a product of Abeyy. Having two recognition engines is quite useful, in that one can improve the results of a scan by determining which of the respective engines provides the most accurate rendition of a given document.

Kurzweil1000 allows the user to not only create bookmarks in a document, but to annotate them with text, thus allowing the user to create their own "margin notes" for a document. This feature when combined with the document summarization make Kurzweil1000 not only a good tool for scanning and reading, but an excellent study assistant as well. One feature particularly interesting, especially to an expert user is Kurzweil’s ability to scan a page without actually switching to the program. It accomplishes this by assigning a hot key that the user can press from any place to scan the next page on the scanner. This allows a user to scan while performing other tasks on the PC. However, though the current version of Openbook Ruby does not contain nearly as many features as does the current release of Kurzweil1000, there are many who would never use this vast array of features. Optical character recognition is still the most important factor to be considered when choosing these products, and as in the past, the results of this year's Dueling Scanners session are far too close to call. Therefore, in order to make an accurate assessment of these products, one must consider what features will be used by a client. Both programs are equipped with a spell checker, a dictionary, and a thesaurus. It should be noted that the Kurzweil1000 spell checking function does provide one feature that we feel is essential if such a tool is to be used effectively, and that is the ability to read misspelled words in context.

Openbook Ruby provides a set of low vision functions that we feel slightly exceeds those furnished with the Kurzweil1000. Therefore, it might be important to choose one product over another when considering the needs of the low vision user. However, the user may well appreciate such things as Kurzweil1000's ability to automatically apply a set of corrections to a document, and the ability, in version 5.0, to have multiple corrections files that can be used with different documents e.g. one for Spanish documents, and one for English documents. Such a feature can also be quite useful for those in professions where a good deal of "jargon" is routinely scanned, e.g. when the product is used by persons employed by government agencies and other professions where what is scanned may not appear to be actual words, when in fact there are no errors. In order to make a final decision about these products, one must determine what the needs of the individuals are, and what type of documents they will need to read. Once having determined the type of documents that will be required by a client, the next step is to attempt to scan and read the documents in question, using both of these products. Since both products offer a "demonstration" version of their software, we urge any and all prospective purchasers to acquire such a demo and put the software through its paces. Only after such a detailed and careful comparison can a good decision be made. Both products carry the same price tag, $995, (although if Kurzweil1000 is purchased with the optional Dectalk Access32 software, the price is somewhat higher) and both products do a similar job when it comes to recognition. We have attempted in this document to mention as many features and shortcomings as we could, the rest is up to you, those of you who will be making purchase decisions for yourselves or your clients. Compare the features included with the products with the features that you need, be certain that you are able to scan and read the kinds of documents you will need, and then purchase the product you feel will most fit your needs or the needs of your clients! Though we have not declared a clear winner, we feel that we have, as was our goal, provided a good place to start, the rest should be up to you!

Below, please find the responses from the two vendors whose products participated in Dueling Scanners 2000. We have asked them to comment or expand upon any issues raised in the above report. They have been asked only to comment upon their own system and not to attempt to elaborate on conclusions drawn about the competing product.

Dueling Scanners 2000 - Response from L&H Kurzweil Educational Systems.

Just a few comments:

1. While timing issues were not measured at Dueling Scanners, it may interest readers of this report to know that we timed Kurzweil 1000 using both the Epson 1200u (which we used) and the HP 5200 that Arkenstone used. We measured a scan time of 10 seconds using the Epson and 21 seconds using the HP.

2. With regards to speech synthesizers, the report states that Kurzweil 1000 automatically detects the presence of any SAPI compliant TTS engines one may have on a PC during installation. Actually, Kurzweil 1000 determines which SAPI or DAPI speech synthesizers are installed each and every time you run the program. The customer can then select which speech engine or engines to use within the product itself. We believe this to be an easier approach than that employed by Openbook.

3. The report makes reference to the absence of a manual on cassette for Kurzweil 1000. Actually, we do have a product tutorial on tape that is shipped to all new customers. Since our manual is very large, we found the better solution was to include the manual online with bookmarks placed wherever headings or new chapters occurred. A beginner can get started with the taped tutorial, and then refer to a vertical list of topics in the on-line manual when they need more detailed information.

4. The reporting of how the table document was handled by the respective products was not complete. We noted, as did others in the audience, that Openbook stopped reading half way through the table and announced there was no more text left to read. Furthermore, it should be noted that Openbook was not left to read the table continuously; rather it was being manually operated to speak word by word. Kurzweil 1000 on the other hand read the table continuously with row and column header information being announced for each cell being read. Our new table feature also allows the user to navigate in any direction through the cells of a table, and speak the appropriate header information at any point. 5. In the section regarding multi-language support, no information was provided as to the number of languages supported by Kurzweil 1000. The present release of Kurzweil 1000 supports 58 languages for recognition, and includes text-to-speech (TTS) for 8 languages. This compares to the quoted 14 languages and 6 language TTS quoted for Openbook. This concludes our comments on the report. In closing, we would like to thank the judges for their hard work writing up this well written, detailed report. Sincerely,

David Bradburn and Stephen Baum

L&H Kurzweil Educational Systems Group
Burlington, MA

Arkenstone Dueling Scanners Response
2000
Jim Fruchterman

As always, it was a delight and an honor to participate in the Dueling Scanners session at CSUN this year. The efforts of the judges and organizer are greatly appreciated, as are their efforts to provide a fair forum for the exchange of important information about the latest innovations in reading technology for people with visual impairments. Arkenstone takes seriously its responsibilities to abide by the rules of the session. In particular, Arkenstone followed the agreed-upon equipment guidelines by bringing the HP scanner that was specified by the organizers, rather than one of the faster scanners we support. We are continuing to follow those guidelines by complying with the judges' specific request to focus our remarks solely on our own products.

Arkenstone is exclusively focused on the success of our users with our reading products. Our goal is for every user, novice or experienced user, to sit down with our product and have high quality, high productivity access to printed material. We want our users to press a single button and have our system do the right thing automatically. We believe that potential users trying out their own documents will generally find that our Open Book works great for them in all the areas they care about: ease of learning, accuracy and speed, quality of interface and features, and service and support. From our inception, we have always made the same recommendation: try out our systems for yourself. We encourage potential users download or request our fully functional evaluation version. I think that dedication to our user explains our continuing track record as the number one reading system in the world.

The report does a good job of communicating our strengths. On real world documents, Open Book does a great job. Our unique hybrid recognition engine does a great job of recognizing pages quickly and accurately. Our user interface offers both a simple keypad interface for beginners as well as a Windows-standard interface for maximum usability. We include the full manual on tape with every shipment because that is what most users tell us they want. We also have documentation in large print, in Braille, and on disk.

In the area of low vision features, we'd like to point out that Open Book's Exact View is useful not only because it provides a picture of the original page, including all graphics and original layout, but especially because the Exact View is linked to the Text View. This means the low-vision user can switch back and forth from Exact View to Text View and not lose his/her place. Both views are synchronized to provide simultaneous visual and verbal feedback. Each word is highlighted as it is spoken. Both views can also be magnified without breaking the link between them. This feature is also useful for trainers and teachers.

In the area of multi-lingual support, Open Book has been fully translated into more than ten languages. Our international users appreciate Arkenstone's commitment to providing them the same high quality product in their preferred language that English speaking users experience in English.

Our commitment extends to the future as well. I'm personally very excited about the new innovations we're working on for the next generation of Open Book. My commitment, and the commitment of the Arkenstone team, is to provide the best and most usable reading systems in the world, as measured by consumers. I invite you to test that commitment personally!

Jim Fruchterman
President, Arkenstone, Inc.


Go to previous article 
Go to next article 
Return to 2000 Table of Contents 
Return to Table of Proceedings


Reprinted with author(s) permission. Author(s) retain copyright.