Registering and logging in removes this ad.
Registering and logging in removes this ad.
Can DNS do this (see post body) in a command form?
Submitted by Mitanoff on Tue, 08/19/2008 - 17:44.
Please forgive my lack of knowledge of this product. I have to submit a discussion paper BEFORE testing the product.
Can the following be done:
- create a voice command that will allow a phrase such as "Target Location" which tells DNS to only select words from reduced vocabulary (i,e, only numbers)
- similarly, I'd need a phrase like "Target Vehicle" with a reduced vocabulary of vehicle types etc. . .
I am not concerned with training DNS to recognize the voice as different users will be using this. I'm hoping there will be increased accuracy with a reduced vocabulary.
- Also, can a macro, once created, but used globally?
I don't need to know exactly how to do it just yet, just if it can be done.
Thanks
Mit
DNS Pro
NaturallySpeaking
Pro 10 is multi-vocabulary capable which would allow you to create a really
small vocabulary but NaturallySpeaking isn't really designed to work
universally with all voices (untrained). Limited speech recognition applications,
such as the programs that allow you to change your car radio station or control
your home environment, are better suited to this task. DNS 10 features a medium
and general-large vocabulary with the general-large vocabulary being about
150,000 words and the medium-sized vocabulary being about 75,000 words. No
doubt you are looking to work with a much smaller vocabulary but the reason why
we mention this is because we experienced no accuracy gains when using the
medium-sized vocabulary. We have also experimented with our own 2000 word
vocabulary in DNS 9 and as amazing as it might sound, our accuracy was quite
bad. NaturallySpeaking is more ideally suited to continuous standard dictation rather
than what you're looking for so DNS may not be the right product for you.
Although you can create application and even Windows specific macros in DNS, the
default setting is global so you will be happy with that aspect of the program but
keep in mind that only DNS Pro is natively macro capable. Preferred
10 is not macro capable but you can remedy that problem with a Third-party
Command utility.
Lunis
Orcutt - Developer of KnowBrainer
&
Host of the
http://www.TheMicrophoneStore.com
A Nuance Gold Certified Endorsed Dragon
NaturallySpeaking Partner/Trainer
ALWAYS Ask If Your Speech Recognition Partner Is
Nuance Certified
NUmbers are important
Thanks for the reply.
Perhaps I should expound upon my intended application.
Soldiers doing surveillance in Afghanistan need to fill out reports upon seeing something "interesting". Typically they type this up. There is a small section (think 1 paragraph) that is used for a description of the event that would not have been covered in the form filling part. The majority of soldiers actually have little typing experience, in addition, while typing (chicken pecking), something could change during the event while their attention is elsewhere. This is undesirable.
I have been tasked to see if STT would help the situation. Even a 70% accuracy rate is acceptable (soldier would review the paragraph later to correct) as long as the LOCATION and TIME are 100% correct. Which is why I had hoped that I could create a verbal command which would tell the S/W that anything said after the command is a number, thus (hopefully) increasing the accuracy.
I currently have PROFESSIONAL V9.0. I have a time line in which to complete my report so I do not think I can wait for V10.
Based on my requirements, what S/W packages would you say is best suited to my requirements? How would you go about increasing the accuracy of the location and time fields?
(I am aware of the Military solutions, i.e. Dynaspeak through Speech@SRI and Adacel. This would be different part of the project)
Mitanoff wrote: Thanks for
Thanks for the reply.
Perhaps I should expound upon my intended application.
Soldiers doing surveillance in Afghanistan need to fill out reports upon seeing something "interesting".
I would think STT is the wrong technology for this. The technology is just not far enough along to allow this to work well.
Perhaps something along the lines of a digital recorder that puts on date and time stamps would be better for this situation.
admin wrote: Mitanoff
Thanks for the reply.
Perhaps I should expound upon my intended application.
Soldiers doing surveillance in Afghanistan need to fill out reports upon seeing something "interesting".
I would think STT is the wrong technology for this. The technology is just not far enough along to allow this to work well.
Perhaps something along the lines of a digital recorder that puts on date and time stamps would be better for this situation.
While I'm uncertain what "STT" means (speech transcription technology"?), Skip's point seems spot on: digital recorders are cheap and reliable, while STT is a wildly experimental reach for a combat situation. OTOH, given how indiscriminately and unaccountably the billions have been slathered around amongst defense contractors recently, I don't see any harm in throwing another one or two billion at furthering the state-of-art for SR.
Bruce
I appreciate all the
I appreciate all the responses thus far. Thank you.
Most of the report to be converted from "Speech to Text" (STT) would be general descriptive text. For e.g.:
" Man in tan coat entering meeting area. Hands over suspicious package to older man in red tunic" etc . . .
Doesn't even have to be spelled correctly as most of the time the soldier's spelling is far from accurate even if it was typed! The information has to be in a written format and has to be uploaded in a timely manner, i.e. within minutes of the event unfolding. The reason for using STT is to make the soldier's life easier. He usually hates typing and doesn't do it well. It is not being used for navigation or combat (Heaven forbid). The only information that needs to be accurate is the NUMBERS. This is not a billion $ project. This is just dipping a toe (a Canadian toe at that! (i.e. no money) into the ASR waters. And it's not even my big toe.
My actual question was sort of addressed but not actually answered.
What can I do to increase the accuracy of DNS professional V9.0 (or 10 I suppose) that will allow any number spoken to be more accurately converted into text.
Is this not possible at all?
Say "switch to numbers mode"
Say "switch to numbers mode"... And combine with GPS co-ordinates from somewhere, rather than saying the location?
Thanks PerH
Exactly what I needed!
Thanks.