Speak to the Dragon

By ADAM GIFFORD

When Police Minister George Hawkins wants to breathe fire, he has a dragon on his desktop to help him. That's dragon as in the speech recognition software Dragon Naturally Speaking.

"I took it up because I had a stroke in 1992, so I am not the best of typists," Hawkins says. "I use it mainly for the stuff I need to do as a local MP - I have a huge staff to do most of the written stuff I do as a minister. I am making notes for myself, so everything doesn't need to be perfect."

Speech recognition has long been in the category of "tomorrow's big thing" for office automation, but never arrived outside of niches like law offices. Even the entry of giant IBM into the field with its competing ViaVoice product failed to set the market on fire.

David Shepherd from Dragon distribution agent VoicePerfect says several MPs have adopted speech recognition systems, and it has started gaining acceptance in fields outside law and as an aid for people with disability.

"I believe it has come of age. With the amount of memory and CPU power that standard PCs now have, it can deliver productivity gains for anyone in document creation," Shepherd said.

"It can be coupled with digital dictation recorders so you automatically transcribe from the recorder.

"There is no longer any need for a touch typist. That frees up the PA to add value to the documents, not just input the raw data."

Barrie Keenan from agri-business consultancy Nimmo-Bell says when a typist-receptionist left the firm, she didn't need to be replaced because the six consultants had moved to dictation software.

"I have been very impressed with the latest version of Dragon, compared with the 1990s versions where-one-had-to-speak-fairly-precisely-and-clearly-one-word-at-a-time," Keenan said. "This is magnificent."

Nimmo-Bell signed up for the business-speak dictionary, which includes standard business terms. There are also dictionaries for specialist industries and for New Zealand vernacular, including common Maori words.

"VoicePerfect did an analysis of a range of our documents, and we are working with them on the farming and agricultural science expressions we need," Keenan says.

The painstaking process of "training" the software to each speaker has been cut to 20 minutes, with the user reading a text which creates a voice file.

This file can be held on flash memory and plugged into the USB port of any computer with the software installed, so it is immediately ready to use.

"It is licensed by the voice, not the machine, so I have it on my work machine, my home machine and a laptop, and I just plug in the memory stick," Keenan says.

Keenan says for one recent job, a 40-minute dictation to a digital dictaphone was transcribed within 12 minutes, with a further 10 minutes for tidying up.

"That turned out be close to 5000 words," he says.

"I have done a lot of dictation, and I'm reasonably good on the keyboard but not totally 'no look'. So if I am referring to documents I want to take ideas from and rephrase or just quoting slabs of text, I would rather do it at 120 to 160 words a minute than my 40 to 50 words a minute typing."

Keenan also uses it for email or for filling in forms, commanding the cursor to move through the fields.

"In this office, we are seeing an advantage with some of our older consultants who came through before computers were widely used at university. They are used to dictating, not typing, and their productivity has leapt up."

Kristi Fitzpatrick from Dragon vendor ScanSoft says sales grew 25 per cent last year, driven by greater usability and accuracy and new customers in the legal, medical and government sectors as well as greater take-up in business.

"You can speak faster than you can type, and if you are also using voice commands to control your computer, there can be significant productivity benefits," she says.

"In Australia we have several customers who use it in call centres to improve productivity and help people who may have had RSI or other problems to return to work," Fitzpatrick says.

What could drive even greater adoption of speech technology is IBM's decision this month to donate key elements of its speech technology to open source organisations.

Its Reusable Dialog Components will be developed by Apache Software Foundation, while the Eclipse Organisation will work on speech editors.

IBM software division senior vice-president Steve Mills said Big Blue wants to encourage the development of standards-based rather than proprietary speech platforms.

"We think we're at an inflection point now, and that all the pieces have come together to a point where the technology is going to sweep across the marketplace and become an everyday part of application deployment," Mills told the SpeechTek conference in New York.

The Reusable Dialog Components include pre-built speech software components, or "building blocks" that handle basic functions such as date, time, currency, locations (major cities, states, zip codes), and could be used in applications such as booking airline tickets over the phone.