The IT industry seems to work like London buses: you see nothing new in a field for ages, then several developments appear in a short time. This time it is speech recognition. Nuance announced the release of Dragon Naturally Speaking version 9; and Microsoft followed up a disastrous demonstration of its own voice technology with an impressive presentation that worked perfectly at the SpeechTEK show.

I have used earlier versions of both products and, if used carefully with good microphones, they can produce excellent results. I haven't had a chance to try out Dragon 9 yet but people who have tried it say it is a big step forward. The accuracy is better and, for the first time, it has a speaker-independent mode. If this feature works well it will open up a huge range of applications for speech recognition.

I must admit that I've always thought that talking to your computer is something to be avoided or, at least, done in private. Microsoft obviously disagrees as its speech technology will be integrated into Windows Vista, and I expect that there will be a good deal of publicity about it when Vista is launched. I hope this doesn't lead to a potentially far worse problem than mobile phones on trains.

It is surprising to me how much research into speech recognition is currently going on; many big companies have teams working on it, as do a large number of universities. If it's not for talking to your PC, what is it all for? The answer may lie in networked applications.

The European Telecommunications Standards Institute (Etsi) has a project called Aurora to develop standards for distributed speech recognition in mobile networks. The idea is that part of the recognition engine runs on mobile devices and transmits speech to the central recogniser. This sort of thing seems much more sensible to me; it's more natural to talk into your mobile phone than your laptop.

Etsi says that one application is to dictate your impressions of a meeting and then have them emailed to you so that you can do the editing when you get back to your office or hotel room. That's quite neat. It's not too hard to think of other neat services that could be integrated with various types of networked servers.

As the popular SpinVox service has shown, it is often much better to receive voicemail messages as text than speech. A good centralised speech recogniser could let companies run a similar service on their networks, especially if speaker-independent systems really are just around the corner. Microsoft's Office Communications Server 2007 should be helpful as it will have integrated speech recognition.

Bill Gates says that speech recognition will enter the mainstream in the next decade; he may be right but it may be mainstream in networked applications rather than for just interacting with your PC. In any case, corporates may now wish to consider whether good speech recognition could improve the effectiveness of their systems.