Author Topic: hal correcting voice recognition talk afterward  (Read 2142 times)

lightspeed

  • Hero Member
  • *****
  • Posts: 6819
    • View Profile
hal correcting voice recognition talk afterward
« on: July 04, 2010, 03:04:46 pm »
I hope Robert see's this and reads it , i don't know if this could be done or not but with voice recognition ( using dragon speak ) when i say something and some words still don't come out right , i try to correct that word but as i am doing it sometimes hal will go ahead and enter it and then answer it . the problem with that is then hal has a sentence with a wrong spelled or wrong word altogether and has learned it that way .
SO IF Robert does read this i am wondering "is their some way for a "smart sense " hal program to "if a sentence isn't right ( as far as context and spelling to do a spell checker and say did you mean ............( and if the answer to the word in question was no , then hal would correct it with the correct word  , if yes then hal would go on and , then answer the response .
I think if something like this is possible in hals programming it would help a lot . [:)]
 

Carl2

  • Hero Member
  • *****
  • Posts: 1220
    • View Profile
hal correcting voice recognition talk afterward
« Reply #1 on: July 13, 2010, 09:32:40 am »
Lightspeed,
  Glad to see we have another user of dragon here, my impression is it's way ahead of the MS version.  I've been using Dragon for quite sometime and have found some ways to increase reconition accuracy quite a bit.  
  I use dragon alone and start a diary about what I'm going to do during the day or have done, helps me plan the day.  I input about 10 text files to dragon.  I also have text about family and friends, cars, electronics, ect.  Very useful is a file about stores, Walmart, Best Buy, street names, and local towns.  Inputing these files right after training really gets the accuracy up.
  Naturaly the mic is important for speech rec. USB input mics will have a higher S/N ratio (amount of siginal to the amount of noise).
The majority of mics on the market have a S/N of 19db using the dragon method of testing. You should be getting around 22 DB.
  I use the Dragon program and the Hal program seperate, Saying "press enter" will input the text after you proof read it or press the enter button.
  Dragon has to spell your word correctly. You increase the chances of Dragon choosing the correct word by inputing the Text files which enlarges Dragons data base.
  I can't speak for Robert, I feel he is busy with the new character and new database for Hal. Hope I've been able to help you with using dragon.
Carl2