Speech regognition for people with difficult to understand speech!
Some people are unable to use off-the-shelf speech recognition software because their speech or breathing isn’t typical. InvoTek’s award winning Supplemented Speech Recognition (SSR) product is different. It is specifically designed for people who cannot use other systems! Years of research went into this product (with funding from the National Institutes of Health) to ensure that it works well for people with speech that is dysarthric (i.e., difficult to understand). SSR tightly integrates word prediction and speech recognition to significantly reduce the amount of typing required to enter text. Our research demonstrated that, even for people with very poor speech, typing requirements were reduced by 65% or more.
The video above demonstrates an earlier version of SR than the product release, but the idea is the same (the toolbar you receive will be more polished). Thanks to Katie and Susan Fager, Ph.D. for the video footage and narration.
daVinci AWARADS 2011 WINNER!
SSR integrates speech recognition and word prediction into a single software program so that you don't have to type so many letters when writing. What makes SSR unique is that it works with difficult to understand speech (often called "dysarthric" speech). Dysarthric speech is usually not consistent enough, nor does it have enough differentiation among spoken words to perform speech recognition on large vocabularies without help. The speech recognizer in SSR is specifically designed to work with this type of speech. We have spent several years developing SSR and received research funding from the National Institutes of Health to make this technology "real". If your speech is hard to understand but it would be helpful to use it for writing, SSR is for you!
SSR uses the first letter of a word to significantly reduce the number of words possible at a given location in a sentence. This makes it easier to perform speech recognition and also enables a larger overall speech recognition vocabulary.
It provides a clear cue for when to speak (e.g. shortly after typing the first letter), making respiratory planning easier for people with severe disability.
It reduces co-articulation among words, making speech easier to understand.
The disadvantage to requiring the fist letter is that typing with SSR is slower than commercial speech recognition software, such as Dragon Naturally Speaking.
SSR learns new words as you say them, so it becomes fine-tuned to the way that you write and the way that you speak. With use the system significantly reduces the amount of typing that you have to do.
Here’s how SSR works: First the user trains the recognizer by saying 240 words three times into the provided desktop microphone (pictured below) so that SSR learns how you speak. This takes a little time and effort, but it makes a big difference on how well SSR can understand you! After training, the SSR will recognize over 500 common words and you can add as many words as you want by just using SSR. When you write, you type the first letter of a word and then say the word. If the SSR knows the word, it will enter it into the text for you. Words that it doesn’t know it helps you type with word prediction. The combination of speech recognition and word prediction reduces the amount of typing that you have to do by 65 - 70%, even if your speech is hard to understand.
The toolbar (illustrated above) is the complete user interface. Picture of desktop microphoneIt sits at the top of your computer screen and ensures that other applications don't get behind it. The first button on the left provides access to all of the user preferences. The SR button provides access to the speech recognition dictionary and gives you dictionary editing control. The Mic button lets you control when SSR is listening to you, and the Man (for "manual") button turns off word prediction so that you can type without help (this is useful when you type web or email addresses). The Undo button makes erasing a mistake easy.
The next 6 buttons are word prediction buttons. When you type the first letter of a word, the word prediction system updates the words in the button list based on a sophisticated language model. Then, it checks the speech recognition dictionary to see which words the recognizer knows. For known words, the background of the button is turned green (so that you know the word should be recognized). When you say the word you want (whether it is in the list or not), SSR checks all the words it knows that begin with the first letter to see if it finds a match. If it does, it puts the word in the sentence for you. If there is no match, it does not insert a word and waits for you to use the word prediction capabilities.
SSR monitors the words it knows and whether it gets them right. It lets you know when it is time to run the speech optimization software to reduce errors.
While it is a little complicated to explain, it is easy to use. SSR comes complete with everything that you need, including a desktop microphone and our USB speech processing module.
Note: The SSR software is computationally intensive and will run best on newer models of computers. Computers that are more than a few years old will run slowly. We recommend computers have at least 2 GB of RAM.