What the Hey?!
Submitted by BruceCyr on Thu, 02/28/2008 - 00:36.
I create a command that uses a bunch of keystrokes to pick a particular menu option in an application. Basically when it gets to the menu it goes down 23 steps. Then I use an {Enter} to pick it. It works for six or seven days, and I'm happy because I save a lot of time and tedious keystroking for this oft-used option.
Then it stops working: No matter what I do, DNS Pro interprets it as text. Switching to Command mode or holding the control key down just produces the familiar <???>. Giving it a new name doesn't help either.
Just coincidentally, the sequence "{Down 23}" that figures prominently in this script does not work in my general repeated directional key script of the form " <1_500>". In fact, no set of values in the range 21-24 work. 20 and 25 do work. I'm sure that has nothing to do with the problem -- it's just an observation for inquiring minds.
So I move on using a work around because I don't have time to figure this out. But today my voice wears out so just for kicks I create a different version of the command which traverses up the final menu rather than down it. I give it a new name. It's working fine. The last bit involves "{Up 17}", which so far has never been a problem.
If I thought Skip would let me get away with it, I'd try to start a pool on how many days the new form works 
If I watched the Sci-Fi Channel more often, I would be inclined to suggest that evil propagates itself through viral programming
Make Your Application Specific Command Global
It's possible that NaturallySpeaking simply doesn't recognize the application from time to time. This can be a problem with application-specific commands and we have experienced this problem on a number of occasions. 1 bulletproof workaround would be to change your command from application-specific to global.
Lunis Orcutt - Developer of KnowBrainer &
Host of the Http://www.KnowBrainer.com Speech Recognition Forum
A Nuance Gold Certified Endorsed Dragon NaturallySpeaking Vendor
ALWAYS Ask If Your Speech Recognition Vendor Is Nuance Certified
KnowBrainer wrote: It's
It's possible that NaturallySpeaking simply doesn't recognize the application from time to time.
Maybe, but then the other commands for that application wouldn't work either, including my new version. But they do, so I still think its some other weirdness going on.
But as a check I reclassified my odd-acting command as "Global" and tried it out. Still doesn't help -- DNS Pro nails the text but refuses to interpret it as a command no matter what I do! In other words it's not a recognition problem, it's an interpretation problem
Thanks for the suggestion.
Bruce
pressing the Ctrl key when
pressing the Ctrl key when you speak the command????
Quintijn
Quintijn wrote: pressing
pressing the Ctrl key when you speak the command????
Quintijn
...is covered in the original post!!!!
Bruce
CTFMON.exe
Plan B.: You are probably already on top of this one but it
doesn't hurt to double check. Can you look under the Processes tab of the
Task Manager to see if CTFMON.exe is running? This utility has been known to
prevent NaturallySpeaking command deployment and you can use CTFMON
Remover to take care of the problem if this is it.
Lunis Orcutt - Developer of KnowBrainer &
Host
of the Http://www.KnowBrainer.com Speech Recognition Forum
A Nuance Gold Certified Endorsed Dragon
NaturallySpeaking Vendor
ALWAYS Ask If Your Speech Recognition Vendor Is
Nuance Certified
Quote: it doesn't hurt to
it doesn't hurt to double check. Can you look under the Processes tab of the Task Manager to see if CTFMON.exe is running? This utility has been known to prevent NaturallySpeaking command deployment
I actually use Process Explorer, but, Yes, I'm cftmon.exe free.
I appreciate the effort -- its conceivable we can figure out what's going on if we dig deep enough. I've been trying a few things too, but I haven't uncovered anything interesting yet.
One notion that's so daunting that I really don't want to approach it is the possibility that mixing legacy .dvc-style commands with VB-style commands (not in the same scripts, of course!) might be causing the problem. Wonder if anyone has any observations on that issue?
Bruce
Mixing Command Shouldn't Be a Problem
We don't believe there should be a problem because we have about 30 legacy.dvc commands and probably about 400 Advanced-Scripting commands. We also utilize HeardWord commands inside of our legacy commands that reference standard Advanced-Scripting commands and we have a handful of VB commands. Everything seems to coexist peacefully.
Lunis Orcutt - Developer of KnowBrainer &
Host of the Http://www.KnowBrainer.com Speech Recognition Forum
KnowBrainer wrote: We don't
We don't believe there should be a problem because we have about 30 legacy.dvc commands and probably about 400 Advanced-Scripting commands. We also utilize HeardWord commands inside of our legacy commands that reference standard Advanced-Scripting commands and we have a handful of VB commands. Everything seems to coexist peacefully.
Lunis Orcutt - Developer of KnowBrainer &
Host of the Http://www.KnowBrainer.com Speech Recognition Forum
Yes, thank you! That's pretty clear.
I have some screwy new data -- what else
First, I commented out the .DVC form of my "[Arrow_Key] [1_100]" script and rewrote it in VB style -- no significant changes there.
Secondly, since the example was right in front of me, I began testing the numeric list, i.e., [1_100]. I documented a subset of numbers whose use would force recognition of the utterance as text rather than as a command. For example, 1-20 okay, but 21-23, 26-29, 31-39, etc. were NOT okay.
That was weird, but not nearly as eerie as the sequel: The subset of non-working numbers changed depending on the length of the list!!! For example, changing the list to [1_80] changed the subset of non-working values. Most of the non-working elements of the former [1_100] list still wouldn't work, but some of them did, while some formerly functional values switched to non-working status!
Most likely this problems is a symptom of something idiosyncratic to my machine, but who knows? And what in the world could it be? And, aside from me, who amongst the 73 or so Pro users who write their own scripts cares?
BTW, in case anyone wants to experiment, the script in question follows:
[arrow_key] [1_N]
Sub Main
SendKeys "{" + ListVar1 + " " + ListVar2 + "}"
End Sub
where [arrow_key] is:
Up
Down
Left
Right
and [1_N] is of the form:
1
2
3
.
.
.
N
BTW #2: Sax Basic bombs when N=>120.
BTW #3 I use square brackets in place of the expected angle brackets.
Bruce
PS: I really am going to go catch some Sci-Fi trash now that I've worked myself into a fine swivet fit
I may even have my annual bottle of brewski.
my experience is slightly different.
I haven't tried all of the numbers you tested, but I tested many of the numbers that give you problems, and the only one that seems to give me problems is 22. Interestingly, when I hold down the Control key, it works. Have you tried training those specific commands, in the Command Browser, such as "Down 22"?
To give you some more data, my command is structured as "[bddeehlr] [1to102]" and is an Advanced Scripting command, with the following code:
SendKeys "{"& ListVar1 &" "& ListVar2 &"}",1
Wait .05
[bddeehlr] consists of delete, down, left, right, and down.
If all else fails, you could break your command into smaller commands.
Matt Chambers wrote:
I haven't tried all of the numbers you tested
Thanks for the suggestions, Matt.
1. I substituted "&" for "+" as a concatenation operator, but that didn't change anything.
2. I wrote an DVC-style version of the command, but that didn't change anything either.
3. I tried breaking the command up into different movement segments, e.g., [1-20], [21-40], etc., but that didn't do anything -- in fact, per my previous results, each segment introduced its own nonfunctioning values.
4. I truncated my original command to the interval [1-20], and then wrote a looping version for the interval [21-200]. On the first try, it correctly interpreted "34", but on every subsequent trial it interpreted values greater than 20 as text.
Gads! Good thing I didn't blow my single annual bottle of brewski ration last night.
Bruce
PS: If I get time, I'll try your previous suggestion in another thread to use spoken/written forms for the command.
PPS: Given result #4 above, there may be something odd about my DNS Pro 9.5 installation. However, I think similar things happened in earlier versions, certainly 8. Both cases involve WXP. I guess I could nail it down if I can replicate the pure-text bias using the spoken/written format.
Matt Chambers wrote: Have
Have you tried training those specific commands, in the Command Browser, such as "Down 22"?
Interesting results: It presents them, but when I say them it says:
"Unable to train' Up 21'. Try training the word(s) individually using the Vocabulary Editor, and provide a spoken form if different from the written form."
Bruce
BruceCyr wrote:
Interesting results: It presents them, but when I say them it says:
OK, I've written a written\spoken form for "[arrow_key] [21-40]" and trained it. It will only train the first 10 values 21-30, but those are nailed perfectly. For 31-40, 31, 33 and 38 do not work. That means that I would have to write interval 10 versions to cover a given span, e.g., 21-30, 31-40, etc. Ho-hum... sounds like a downtime project. I guess I'll just write one when I need it:-)
Bruce
PS: Unless of course there's a way to train commands which don't show up automatically in the dialogue.
N.b., none of this explains
N.b., none of this explains why {Down 23} doesn't work in script! It's not as if I say the phrase!
I am reminded of the following judgment on Nuance's orientation towards its desktop product, which I referenced in a previous thread (http://www.speechcomputing.com/node/1720):
'[A] former Nuance engineer says that "there just wasn't corporate-level support for internal innovation. The focus on growth through acquisition has the unfortunate consequence of shutting down the internal stuff." This engineer, who still works in speech recognition, says he knows of others who have left because of similar frustrations.'
See the original at: http://www.boston.com/business/technology/articles/2008/01/20/new_ways_t...
Bruce
Oh well -- what a vast waste
Oh well -- what a vast waste of time and effort! Breaking the commands up into interval 10 form and repeated training doesn't improve my results. The command capability is supposed to improve productivity, not retard it!
Bruce
What the Hey?! My repeated
What the Hey?! My repeated arrow movement commands are working correctly today -- I guess you could say: Another day, another Dragon
FWIW, I took the time to page through the Dragon Error Log for yesterday and couldn't find any obvious sign of a problem.
Bruce
Some thoughts after having
Some thoughts after having read this whole bunch of messages:
1. the numbers 12, 13, etc present in my experience intermittent problems. So one day they can hit, another day they refuse.
2. It could (hypothesis) be influenced by the sound quality/quality of your speech profile
3. Numbers are caught in dictation in some (for users hidden) module. When using 1-99 in commands, we expect the command can in the same way catch numbers from this "hidden module". But we do not know how this works.
4. You can always add spoken forms to the items in your numbers list, 1\one, ... 12\twelve, 23\twenty three etc.
5. In unimacro I have grammar rules running which catch numbers as eg "20 3" and merges this to "23".
Quintijn
Hi, Quentijn. Thanks for
Hi, Quentijn. Thanks for taking the time to add comments. See my responses seriatim below.
2. It could (hypothesis) be influenced by the sound quality/quality of your speech profile
This comment doesn't really say anything! Are you saying that something in my speech profile corrupts some of the numbers from 21 through 99 but nothing else? To merely state this argument is to refute it.
3. Numbers are caught in dictation in some (for users hidden) module. When using 1-99 in commands, we expect the command can in the same way catch numbers from this "hidden module". But we do not know how this works.
Hidden module! Alright!! Who needs the Sci-Fi Channel when there is a conspiracy theory afoot?
4. You can always add spoken forms to the items in your numbers list, 1\one, ... 12\twelve, 23\twenty three etc.
You didn't read thoroughly enough -- that doesn't work in this case.
Bruce
Hi Bruce, I do not
Hi Bruce,
I do not understand you completely.
2. I do not talk about corrupted speech profiles, but of grammars inside NatSpeak (hidden or not) that are better recognised if the quality of sound/speech profile is better. Just a hypothesis I am working on. In Dutch we have a problem that the Dutch equivalents of "choose n" in the Spell window are sometimes not recognised! Same hypothesis, additional "Train user" session could solve this. But only a hypothesis! I mentioned this, because you report that some time "later" things got to work again...
3. Sorry I might have missed earlier comments, but what is Sci-Fi? How do commands extract numbers from a list, how does NatSpeak know that "23" has the sound of "twenty three"???
4. I did not read that, and it does not solve the problem in my Spell box example - above - either. But in other grammars it certainly does.
Greetings, Quintijn
Quintijn wrote: but what is
but what is Sci-Fi?
SciFi is a television channel that airs on most satellite dishes and cable systems in the United States and Canada. I don't know if it's available elsewhere.
It started by showing SCIence FIction (hence the name) movies and programs. Lately it has progressed from real SciFi into a dismal swirling cesspool of shows that will virally program you into an evil being (or they are trying anyway). It's really gotten bad with few programs to redeem it at this point. I've still not figured out what wrestling has to do with SciFi but they air it anyway...
But if that's the drivel the public wants...
Probably more than you wanted to know...
Unfortunately we get it here
Unfortunately we get it here on the other side of the pond, where it joins the rest of the rubbish that is on most of the 150 odd cabled TV channels that I get
The only way to watch them is to let the mind fall into oblivion, and the eyelids will then drop into a closed position for an hour or so, after which one feels much refreshed from having this rubbish available
Quentin
I've always liked science
I've always liked science fiction writing. They have totally ruined the entire genre as well as the offerings on their cable channel. The only way that I can watch anything on there is to use a digital video recorder. That way I don't watch the commercials and the drivel they have there as well.
Quintijn wrote: Hi Bruce,
Hi Bruce,
I do not understand you completely.
2. I do not talk about corrupted speech profiles, but of grammars inside NatSpeak (hidden or not) that are better recognised if the quality of sound/speech profile is better. Just a hypothesis I am working on. In Dutch we have a problem that the Dutch equivalents of "choose n" in the Spell window are sometimes not recognised! Same hypothesis, additional "Train user" session could solve this. But only a hypothesis! I mentioned this, because you report that some time "later" things got to work again...
3. Sorry I might have missed earlier comments, but what is Sci-Fi? How do commands extract numbers from a list, how does NatSpeak know that "23" has the sound of "twenty three"???
4. I did not read that, and it does not solve the problem in my Spell box example - above - either. But in other grammars it certainly does.
Greetings, Quintijn
Quintijn,
Re: #4: Our esteemed Administrator has covered that ground so thoroughly that I'm ashamed to admit I've ever trodden on it
Re: #3: We have the same problem in the English version. Pro users write their own scripts, say "Pick N" -- I would think you could do that in Vocola/Unimacro too.
Re: #2: Anything you discover will be of interest to all of us -- Good Luck!
Bruce
BruceCyr wrote: [Re: #3: We
[Re: #3: We have the same problem in the English version. Pro users write their own scripts, say "Pick N" -- I would think you could do that in Vocola/Unimacro too.
I have an (unpublished] unimacro grammar that "overloads" "choose n" in the spell window. Never misses a command!
Quintijn
PS I will report if I have more info on my little hypothesis.
PS About that Sci-Fi part: the numbers catching is no Sci-Fi, but something hidden (to us) in NatSpeak. My question about the coupling of this (working in dictation) with command lists in commands is not answered.