As Open Office is Open Source, perhaps someone "in the know" could ask the developers if they could make it "Select and Say" compatible on the Windows version. I wonder if the problem might be in Java and not OO?

I don't know what's involved or I'd ask them!

I tried Dragon with Jarte and AbiWord Portable, and found that they weren't compatible, sorry to say. It's especially perplexing with Jarte, which supposedly uses WordPad somehow.

Apparently I am not communicating clearly.

DNS doesn't care what application you're dictating into or what operating system you're using. DNS is a recognizer, period. By itself all it does is listen to your speech, convert it to a format that the recognizer can interpret, analyze it against your Acoustic Model, analyze it against your Language Models, and spit out the results. That's all it does.

DNS does not communicate directly with any application. That is done through SAPI. If you go to the folder with the speech files (Windows\speech), select Vdict.dll or Vcmd.dll, right-click and select properties from the pop-up menu, and take a look at who owns SAPI. Lo and behold you'll find that it happens to be Microsoft.

All communications between applications and DNS are controlled by SAPI. SAPI communicates with DNS to tell DNS what words are currently visible on the screen in a document. SAPI takes the results of DNS and pastes it into your document or the text window into which you're dictating. Microsoft maintains the standard text edit formats via SAPI that it can use to keep track of words (visible words on the screen) and is responsible for communicating the location of text in a window back to DNS. DNS hands off the results of recognition to SAPI and SAPI writes that information into your document.

Therefore, there are three problems with nonstandard Windows. Please keep in mind that this is all intended tongue-in-cheek, so please don't take offense.

1. SAPI SAPI SAPI SAPI SAPI SAPI SAPI SAPI SAPI SAPI SAPI SAPI SAPI... ad infinitum.

2. SAPI is not equal to Nuance, SAPI is not equal to DNS, SAPI is equal to Microsoft. Write that 1000 times on the blackboard.

3. Nonstandard Windows are those that do not conform to SAPI supported text editing formats. If SAPI won't support it, DNS won't support it. But it's not DNS's problem. DNS relies on SAPI. Therefore, DNS relies on Microsoft the author of SAPI. If an application text window is not SAPI compliant, it is a nonstandard window.

It's time to stop blaming DNS. It's not DNS's problem. If you want to scream bloody murder at somebody, scream at Microsoft, or scream at the application developers for not making their text window SAPI compliant. It's their problem. The only thing that Nuance, or ScanSoft and Lernout & Hauspie before them, could ever do with regard to nonstandard Windows is to try to find the best text editing format with which to communicate with an application text window, by SAPI is still the interface that does the actual communication in both directions. If none of the formats supported by SAPI work in those Windows, or work fully (i.e., Select-and-Say), then there's nothing that Nuance can do to fix the problem. Lernout & Hauspie and subsequently ScanSoft and Nuance have tried the best they can to deal with the problem in the only way possible. That is, via the dictation box.

Chuck Runquist
Former Dragon NaturallySpeaking SDK & Senior Technical Solutions PM for DNS with Lernout & Hauspie

"We are all victims of mythology in one way or another. We are the inheritors, and many times the propagators, of a desire to believe what we want to believe, regardless of whether or not it is true." -- J.V. Stewart

It occurred to me that I left out a point in the previous post. Therefore, to clear up a point, here is how DNS works with applications and SAPI.

1. You dictate something.

2. Through a series of algorithms your speeches converted from analog to digital, from digital input to a voice pattern, from the voice pattern to a phonetic equivalent utilizing the Acoustic Model and the vocabulary (i.e., the underlying IPA -- International Phonetic Alphabet for the specific language vocabulary being used).

3. If you watch the results box very carefully you will see how the transition progresses from the Acoustic Model representation of your dictation (speech) through the comparison to the Language Models for context. In other words, the first representation is the Acoustic Model interpretation of your dictation (speech) based on your enunciation of words derived from either the speaker independent Acoustic Model or your general training (initial training).

4. Once the recognizer has gone through the entire process and shows in the "BestMatch III" top selection match, natspeak.exe, which is DNS's linking loader and not DNS itself because DNS is MREC.dll (the recognizer), sends the end result to SAPI, which then attempts to write the text to the application window. All the other surrounding DLLs and ActiveX controls/COM programs are simply auxiliary functions which natspeak.exe uses to process user requests. This is why there is a delay between the time that you pause and when the text is displayed in the application window.

Keep in mind that no application performs any operation that is controlled by Windows. When you type on the keyboard, Microsoft Word does not insert the keystrokes correctly into your document. The keystrokes are passed to (Oh my God!) COMMAND.COM -- YES MSDOS. COMMAND.COM contains all of the communications protocols required to convert keystrokes into characters. This is called the console, or con for short. The results are then passed back to Windows which displays the results in your document. This is the way it has been done since the beginning of DOS. So, when you talk about applications interfacing with other applications, they only do that through Windows, of which COMMAND.COM and MS-DOS.sys are still a part.

Also, note that the SAPI speech file, even though they are located in their own separate folder, are basically system files. In SAPI 5.3, which Microsoft uses in its WSR in Windows Vista, the SAPI files all written to the Windows system folder(s) (i.e., system32, etc.). The reason that SAPI 4 is separated out in its own set of speech files and folder is because SAPI 4 was originally a system component of Windows 98/98 SE. Microsoft installed SAPI 4.0 when you installed Windows 98/98 SE. At that point in time, Microsoft was not involved in creating its own speech recognition application. SAPI was also not installed in Windows 2000, which came out after Windows 98 SE. Nevertheless, when Windows XP was first introduced, it was introduced in conjunction with Microsoft Office XP, which was Microsoft's inroad into the speech recognition arena, and also why DNS users have problems with ctfmon.exe. Ctfmon.exe is installed automatically every time a user installs a version of Microsoft Office from Office XP up through Office 2007. This occurs regardless of whether or not the user installs the "Alternate Input" option in Microsoft Office.

Nevertheless, the above four basic functions are independent of the operating system, as well as application independent.

As far as communicating with an application, when you load Dragon NaturallySpeaking and activate a user, and then open an application, natspeak.exe, through its various functions as a linking loader, sends a query to SAPI and essentially asks SAPI if there is a text window (document window or corresponding text field) and if so what format is a using. If SAPI in turn says, "Hey I got it. I recognize it as being one of my supported text edit formats.", then DNS applies the appropriate format for dictating into the document or text field (for the sake of common jargon, will call it a text object), and identifies it as Select-and-Say enabled. However, if SAPI says <???>, or in essence "I have no idea, then DNS applies the most logical and most commonly used text edit format. However, DNS identifies such text objects as nonstandard windows. And, if that text object is capable of certain types of SAPI communications, then some or all of the features that are normally enabled with Select-and-Say may or may not work. Note also that Select-and-Say is not simply allowing DNS to communicate with an application via SAPI, Select-and-Say also enables Natural Language Commands in a particular application. This is what the case is with dgnword.dll and dgnoutlook.dll. That is, these two particular DLLs are not simply for the purpose of communicating text back and forth via SAPI, they are also and primarily used for enabling Natural Language Commands for those applications that DNS supports using what are called constrained grammars (i.e., only applicable to that application vs. global commands). DNS does this through a number of low-level SAPI calls, some of which are part of dnsapi (SDK low-level SAPI interface functions) and some of which are controlled directly through SAPI itself. Regardless, the primary SAPI DLLs that are used to communicate back and forth between DNS and an application text object are, as noted in a previous post, Vdict.dll and Vcmd.dll. Vdict handles dictation issues, while Vcmd.dll handles the communication and execution of commands (MSAA -- Microsoft Active Accessibility, or as DNS notes that Active Accessibility, as well as scripting commands).

The point here is that DNS does not connect directly with an application text object except via SAPI. Delete the SAPI files and you'll find out real quick that DNS crashes with the error message that the speech files are missing. I'm sure that everyone has seen either a post on this problem or experienced it themselves wherein the user has to download and reinstall the spchapi.exe. This also occurs if the wrong version of the speech files are installed. Nonetheless, no SAPI no DNS.

Lastly, what capabilities are or are not accessible and/or functional with regard to dictation into nonstandard windows is determined exclusively by whether or not DNS's calls to SAPI and/or SAPI's response to such are available in, for lack of a better term, intelligible. For example, you try to do correction in a nonstandard window and the text is selected, or the correction that is returned, are not in the right place. You go to dictate into a particular text object, and you get nothing even though the results box displays your text. Or, you get <???> And no matter what you do you cannot get the text displayed in the text object window. Also, pausing and then continuing to dictate results inplaced between the last utterance and the new utterance, coupled with no capitalization after terminal punctuation when you move from one point in a document to another to insert text, etc. and so on ad infinitum.

This is not a technical explanation as much as it is an attempt to make the process understood as simply as possible. So, you programmers out there don't grill me as to whether this occurs are that occurs, or not doesn't make sense. The explanation is an explanation of the process, not the methodology or the technical details. If I've overlooked something in attempt to keep it as simple as possible, I apologize. Nevertheless, this is basically how it works and where the necessary "fixes" must be made in order to make a nonstandard window work properly with DNS. On the other hand, yes, it may be possible for the DNS programmers to figure out a way around this problem. The chief Belgian engineer for Lernout & Hauspie was adamant that he could make L&H VoiceExpress work with Java, but he's long gone. Nonetheless, possible or otherwise, it won't be easy and it won't be soon, if at all.

Chuck Runquist
Former Dragon NaturallySpeaking SDK & Senior Technical Solutions PM for DNS

"Facts do not cease to exist because they are ignored." -- Aldous Huxley