speech recognition software – Voice Xpress version 5

From Lernout and Hauspie

Don’t underestimate how much the technology has progressed – if an early brush with speech recognition software might have raised an eyebrow, a few years on and a serious look at Voice Xpress version 5 might make an eyeball pop. 

Working at speeds of 130 wpm – more than the fastest typists, watching this at work is seductive. Unlike ‘discreet word’ recognition packages that process single words and work with less powerful computers, Voice Xpress uses the likelihood of one word following another to generate impressive results. Its ‘thinking’ in context lets you talk naturally pausing only to say ‘comma’ and ‘full stop’.

The fantasy of dictating a school report say, as you put your feet up and talk to the computer seems almost realisable. You talk as fast and as clearly as you can while the machine types out the words in swathes. The reality is that you correct errors as you go but you can do this in a way that is entirely hands free. You’ll say for example, ‘correct realise of all’ and spell out its correction (realisable). Dialogues may appear with alternatives like ‘to’ or ‘two’ and you say which you want. You train it for any specialist vocabulary or acronyms so that for example, ‘ICT’ is no longer returned as icy tea.

There’s clever machinery behind this. It surprises by making sense of phrases like ‘you need a 200MHz Pentium’ and putting this into the right shape. It can act a bit too clever and interpret ‘from five to ten’ as ‘from 9.55’. Persistence teaches its canny ways, and there’s a way of speaking that works and others that do not. My problems were outside the realm of reasonable complaint: I could not think fast enough for a machine that prefers clauses, nor does it do well on tongue slips, dry lips or an incurable chewing gum addiction.  

You must first train it for your voice and this takes under ten minutes. This makes a profile you can keep if you move to another machine. As you’d expect the voice transcription happens in whatever application you use. For example, you talk straight into Microsoft Word, Excel or even Windows Notepad.

The package comes in flavours, some offering extra voice commands listed on-screen but less may be best. You can control most programs or say “bold that; make it red; bullet that”. When used to transcribe tape recordings (by voicing the words into the mic) a mixture of talking and mouse clicking is more efficient. Nice as it is to work with feet up, it’s easier to click things than mouth words.Version 5 offers to filter out ums and ers; as well as lots more voice control. Best is Lernout & Hauspie’s Realspeak that reads the screen, in a human sounding voice. Hear this, it may well be the sound of tomorrow’s bedtime story.