what is best vocab size?

hi

can anyone tell me if the 150,000 word vocab you start with on dns9 pro can hamper word recognition? I wondered if a 5,000 word vocab, might be better?

gav

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Chuck Runquist's picture

Don't mess with the vocabulary

gavPWD wrote:

hi

can anyone tell me if the 150,000 word vocab you start with on dns9 pro can hamper word recognition? I wondered if a 5,000 word vocab, might be better?

gav

gav,

Don't play around with the vocabulary unless you know what you're doing. The vocabulary is the key behind recognition accuracy. A 5000 word vocabulary doesn't contain sufficient words necessary to give you proper accuracy. You need to create a vocabulary that consists of at least 50,000 words and the words that must be contained in the vocabulary have to conform to normal structured writing. Otherwise, your accuracy will go to hell in a handbasket (i.e., 99.9% down to about 20%) if you do it wrong. KnowBrainer has created a 50,000 word vocabulary that seems to work fairly well, but still has some flaws. Creating a smaller vocabulary may, to a certain degree, improve performance, but not necessarily accuracy. The only advantage and accuracy from a smaller vocabulary is that there are fewer words that are what we call speech homophones (words that have a similar or identical underlying IPA (International Phonetic Alphabet) phonemes, which is what all speech recognition uses to assign pronunciations to various words).

What I would recommend doing is that when you find a misrecognition that uses a word in the vocabulary that you will never use, simply delete that word.

Nonetheless, you can really mess up your user profile by messing around with your user vocabulary if you don't know what you're doing.

DNS works just fine, albeit that it requires some training and correction from time to time, with the 150,000 word vocabulary.

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

"Life's Rule #1: Once you pull the pin, Mr. Grenade is no longer your friend." (Variant of Murphy's Law - Edward A. Murphy, Jr)

Chuck, Doh! I get it, I

Chuck,

Doh! I get it, I think -- if you say something that's not in a reduced vocabulary, the recognizer will work hard find SOMETHING to attach to the sounds it heard, which could lead to compound recognition errors. Something like that?

Bruce

Thanks Mr Runquist

I get it..

My rec. acc. went down with my 1500 word vocab, and I, as a newbie, had expected the opposite.

thanks a lot
gav

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.




view recent posts