Words out of your mouth
Tatum Anderson draws on her experience as a user to help newcomers get the best out of voice recognition software
It’s inside smart TVs and cars. It’s behind intelligent personal assistants within millions of tablets and mobile devices.
Banks use it to identify customers. It can even be used to order pizzas. Software that recognises the voice and converts it into text, or voice recognition, has gone mainstream.
For disabled people, this software has long been an invaluable tool for those unable to type, use a mouse, or write.
Chris Wood, who was left tetraplegic after a bike accident, began to use voice recognition software to help him return to work as an automotive consultant.
“I would not be able to use a normal computer without voice recognition computer software. The computer would be absolutely useless to me,” he said. He cannot move his hands and fingers but is able to draft documents faster than he used to manage with a keyboard.
And a 2011 study by Georgia State University found that voice recognition software could improve the writing fluency of students with conditions including spina bifida, cerebral palsy and muscular dystrophy. Students doubled, tripled, or even quadrupled their writing rate using the software.
Most use Dragon NaturallySpeaking, by US company Nuance, on their PCs. Home, premium and professional versions of Dragon Version 13 range from £79.99 to £279.99. Options exist to buy volume licenses or add specialist terminology, such as medical libraries.
Text can be dictated and, most importantly edited, using commands such as “Delete sentence”, “backspace” or “insert before comma”. Users can correct words, if Dragon gets them wrong. It offers five similar alternatives to choose from.
Documents can be scrolled through and punctuation inserted. A text-to-speech feature helps blind users listen to what they’ve written, identify and correct errors.
Most importantly, saying, “wake-up” or “go to sleep” starts and stops recognition by voice. That helps users unable to press buttons on a keypad. And a dialogue box, or Dragon bar, sits on the screen displaying text the software has recognised.
Dragon learns too; it creates a user profile containing unique speech patterns and word choices for a particular voice.
To this profile are added new terms– from acronyms to medical jargon (there are options to train the software to recognise unfamiliar words, called custom words). It analyses emails and stored documents to improve accuracy. Users can make up their own commands too.
This signals Dragon to insert standard pieces of text, such as home and work addresses or standard letters.
And because Dragon is integrated with third-party applications, such as Microsoft Office, it’s possible to manipulate emails within Microsoft Outlook, spreadsheets within Excel and search the web via browsers from Microsoft Explorer to Mozilla Firefox.
Switching between applications is possible, as is posting to Facebook and Twitter. Skype calls can also be made and answered using voice recognition (although Dragon must be switched off during the call).
Managing and transcribing recordings of dictation can be done from several third-party applications such as Sony’s Sound Organiser. This software features a Dragon button, to enable recordings to be transcribed.
Transcribing recordings is a key requirement for many disabled users. 23-year-old Oscar Robinson was diagnosed as severely dyslexic while at university and transcribed notes from lectures using the software.
By listening to recordings of his lecturers he was able to simultaneously dictate their speech verbatim into his voice recognition software. “Trying to remember what I’ve read on the page and spell it correctly onto another page was a nightmare. If you can read and write it at the same time, it’s much better,” he said.
Dragon works better with some applications than others, however. Some versions are able to insert text within OpenOffice documents, for instance, and others do not. Webmail, too, is hit and miss.
And Dragon only works really well when you know the commands. Unfortunately, the number of voice commands runs into pages and can be difficult to remember.
Those who persist, however, generally see better accuracy. There are also are companies who supply specialist trainers to help users get used to software.
Sometimes, the software doesn’t work well at all. It does not respond well to hesitation, stammering or a lot of background noise, for instance.
And Dragon can be achingly slow. It is memory-hungry and can challenge some PCs. Wood says his desktop computer can sometimes grind to halt.
“The profile I have built up now is so large, that it takes a lot of RAM to run it and a lot of memory for it to run smoothly,” he said. “The computer freezes and you have to shut down and restart. I know a few people who have come across that problem.”
And the Georgia State University study found that users with nasal problems - users with Duchenne muscular dystrophy who can manage a few words between breaths - reported high error rates.
They spent so much time correcting the errors that researchers thought the software might be better for first drafts only.