Voice Recognition For The Visually Handicapped?

MrIcee asks: "I was recently contacted by an individual who is extremely visually impaired. He wishes to be able to user voice recognition in his PC to dictate letters and to control his web browser and other software. In examining the available software (such as Naturally Speaking, etc) all of them require a fairly involved training session. The problem with training is that they tend to display paragraphs for him to read - but he is unable to see the words clearly enough and fast enough for proper training (even though he can use JAWS for reading screen text). I have seen references to Kurzweil VOICE software, but it seems dated and there is no good indication that it will work as promised. Does anyone out there have experience in voice recognition software that requires minimal or no training?"

Comments

Dragon..

[zulux]

The version of Dragon that I'm familiar with uses plain old text files for the imput for it's training session. I'd imagine you could replace them with somthing that it also avaiable in brail.

Voice recognition for text requires training

[Chris Canfield]

I'm sorry, but the current state of the art in available voice recognition (Dragon, of course) is pretty unreliable on its own. With several hours of training, it gets about one in twenty words wrong. It also doesn't understand new words and many proper names, like Kazakhstan, Khomami, Schwarzkopf, etc. As long as he only speaks to people named Greg and doesn't talk about the 60 Palestinian shops burned near Tulkarm, he should be OK. Otherwise, he will have to recognize that the imputing software made a mistake (which, as you said, is unlikely), highlight the text, and replace it manually with a keyboard. And without training, that accuracy drops precipitously.

Why do all of that? He's vision impaired, he should have a friend teach him to touch-type.

Browser / Program control is easy. Get a Mac. They have had built-in voice-based program control with speakable items for about ten years now, and there are products that extend the existing control base.

The google directory has links to a lot of speech recognition companies, but I still think your client would be most happy with OSX's recognition options. It doesn't have to be trained for a voice, though its full power does require some Apple Scripting. And there are many products that build on apple's relatively solid base, if you choose not to script "go back" to "apple-B" in the browser yourself. Having used it myself in the past I can confidently say it exists, and used to work at least most of the time.

That's more than can be said for most voice-recognition products. Bring him to an Apple store, and have him try it out. If you go to CompUSA or another superstore, bring your own apple guru to show you.

Re:Stick with Naturally Speaking

[cornice]

The issue is that the software is trained for the specific voice. Even if someone is there to help, the phrases have to be spoken naturally by the person who will use the software. This process can take many hours to reach any kind of accurate training. It seems that it could be possible to memorize all of the training phrases and use the help of a friend for seeing the start and stop cues but it's going to be a long tedious process.