This is the vision of the future being promoted by Responsive Systems, the British licensee for DragonDictate, an American program that applies computerised voice recognition to standard desktop PCs. Voice recognition has long been technically possible - in laboratory conditions. The company's aim is to turn it into a familiar feature of the British office. Judging by the present software, it may succeed, but not in the five years in which it hopes to achieve this.
Voice response systems have come a long way, but DragonDictate still has limitations. The program allows users to dictate, instead of type, into a wide range of standard PC applications - most importantly, word processors such as Word for Windows. Depending on the option you buy, the program is 'ready-trained' to recognise a vocabulary of 5,000, 30,000 or 60,000 words.
Not quite ready-trained, in fact. As originally written, the software was better at understanding accents from Brooklyn than from Birmingham. Dragon's UK subsidiary has therefore devoted considerable effort to retraining it, using people who live near its Cheltenham offices. If you don't have a Gloucestershire accent, you have to work a little harder.
Although you can start using the system immediately, the advised procedure is to read a list of 100 or 300 words to it, so that there can be no mistake about the basic commands, such as 'delete' or 'go to' required to operate a word processor. If the computer is happy with your diction, then you can start dictating. If not, you have to repeat the word until it is.
With familiar words and punctuation, enunciated at a steady and monotonous 30 words a minute, the system is adequate. It can be a little unsteady with rarer words, and has trouble understanding users with colds. The program provides a safety net, offering alternative spellings for words whose spelling it cannot guess from the context. Its recognition feels equivalent to that of a slightly temperamental but largely loyal dog.
Proper names are more of a problem. To teach the program to recognise my name, it was not enough simply to say 'Philip' and spell it out. The computer had to hear each letter as if it was an airline reservation code: 'Papa, hill, India, limerick, India, papa.' It seemed to find French words easier than English.
Like similar packages, DragonDictate works by recognising the sound patterns produced by the speaker and gradually building up a voice file of his or her personal vocabulary. 'The system works on a phonetic level, and so it trains itself on all similar sounds,' says Dr Melvyn Hunt, research director of Dragon Systems UK.
William de Marvell, the company's education director, claims the system is 'only as good as the user'. But it is too convenient to blame the operator for mistakes. Natural human speech is unpredictable and idiosyncratic; that is one of the things that distinguish us from Daleks. If speech recognition systems are to supersede dictating machines and typists, they will not be able to demand consistency of pronunciation or diction.
As it stands, the system is better at producing drafts than at turning them into finished documents free of errors. The 60wpm claimed will probably be achieved only by dedicated users who are willing to put in hour upon hour to train the recognition system, and who can teach themselves to talk in ways it understands. Others may well find it less trouble to learn to touch- type. None the less, the program will undoubtedly be a boon to disabled people who would have great difficulty using a keyboard, and for whom 30wpm is at present an ambition beyond reach.
For all its limitations, Responsive Systems says the program is selling 10 copies a week, mainly to 'the inexperienced, the unwilling and the unable'.
Buyers include an oil company, lawyers and writers, and the company has even sold one system to the Vatican.
Responsive Systems sees the greatest potential among business people and professionals, and among those worried by keyboard- related injuries. 'Seven out of 10 users do not actively use the keyboard. In business language, you typically use between 5,000 and 8,000 words - so our systems are ideal,' says Rupert Christie, a systems analyst with the company.
Basil Preuveneers, a partner in a south London law firm, is one of a small but growing band of users. He uses voice dictation for letters and documenting legal work, and describes it as miraculous. 'The knock-on effects,' he says, 'are quite frightening.' Once such systems become common, secretaries will leave their keyboards and be given more interesting work.
Meanwhile, Responsive Systems is developing further packages using the technology. It is running one project on interactive learning, and is working on voice recognition through radio transmitters. The system has worked successfully with FM radio over a distance of a few hundred metres.
Specifications: the programs will run on a 486-33 PC, but require 12M of RAM. Prices range from pounds 600 for the 5,000-word vocabulary version to pounds 1,700 for the 60,000-word version.Reuse content