Available on the PC platform for the past two years, IBM’s ViaVoice is the first continuous-speech program to be available on the Mac. Boasting a UK English vocabulary of 100,000 words, garnered from analysing one billion words of text, ViaVoice also allows for a user dictionary of up to 64,000 words.
The initial voice-training is like having teeth pulled: painful, but necessary. Setting up the mic level and background noise – followed by a voice analysis – takes up to 45 minutes. At least we get something classy to read – I’d forgotten how enjoyable Alice in Wonderland is.
Because we are idiosyncratic when it comes to writing style, ViaVoice can analyse existing documents. This helps it build up a model for added accuracy.
Dictation takes place in ViaVoice’s SpeakPad, a word processor similar to SimpleText. The session starts with a brisk “what can I say” which brings up a window of the same title that lists all spoken commands: cursor movements, dictation commands and text selections.
The first thing to emerge as you start working is how invisible the system is. Dictating at around 120 words per minute, pretty fast by anyone’s standards, it’s not difficult to knock up a 95 per cent rate of accuracy – that’s around half a dozen incorrect words a minute at worst when dictating at speed. The correction procedure is equally impressive. Say, for example, the word “invisible” in the first line of this paragraph came out as “visible”; saying “correct visible” brings up the Correction window with a selection of possible alternatives.
If none of these fit the bill, the correct word can be spelt in, either letter by letter or phonetically, in alpha-bravo form. The What Can I Say window even changes to show this set of words. Another way of making this correction is to say “select visible”. ViaVoice highlights the word, allowing you to utter the replacement word “invisible”.
There are macros as standard for punctuation purposes. Saying “fullstop” results in a “.” followed by a space. In the same way, oft-used phrases can be recorded and then entered via a keyword. For instance, email addresses will always be awkward to dictate. Pick a word you never use in text and record this as the macro keyword. It will then appear in the What Can I Say list.
Text can be styled and formatted and then transferred to AOL, AppleWorks, Microsoft Word, Netscape or Outlook Express via a simple voice command. For other programs, the standard copy and paste has to be used via the Mac’s Clipboard, with the loss of any formatting information.
OK in the UK
Rather than simply release a rehashed US edition in the UK, IBM has worked hard to present a thoroughly usable version of ViaVoice. The UK vocabulary is impressive – football teams, including Tottenham Hotspur, towns and counties such as Gloucestershire and Bicester, and TV entities such as the Teletubbies are all recognized. Similarly, UK phrases with odd homonyms are well catered for. “I practise golf in my law practice” and “licensed to kill with a gun licence” both work. Also, difficult phrases such as “check whether the weather has improved” and “please listen to my pleas” are handled impeccably.
Another UK oddity is the formatting of currency, time and dates. “September the 21st” results in September 21st while “thirty five pounds and 60 pence” produces £35.60 as expected. ViaVoice supports the most natural way of saying number-based terms.
The minimum system requirement is a 233MHz iMac with 64MB usable RAM, 300MB available hard disk space and Mac OS 8.5.1. Most standard iMacs will need a RAM upgrade to at least 96MB, allowing for system usage. IBM also advises the use of 10MB of virtual memory. Our testing was done on a G3 PowerBook with virtual memory turned off – with no appreciable loss of performance.
IBM has had three years of feedback on ViaVoice from the PC market so what we have is not your usual version 1.0 program; it’s polished, highly functional and easy to use.
The UK release is a version 1.03 and incorporates fine-tuning on the handling of background noise, as well as faster text analysis. It’s good value for money – and if you spend lots of time at a computer it’s for you.
One gripe is ViaVoice’s US-only version of the suffix “-sing” (correct UK useage is “-zing”). It would be useful to have a user preference for either ending. By the way, the Andrea low-noise headset comes complete with plastic snap-ons for each iMac colour.