'Did you say Harriet?' asks the voice-dialling function on my phone in its calm Japanese-American accent. 'Henry,' I repeat, firmly.
'Did you say Helen?' the voice asks. I repeat myself. 'Did you say Herb?' 'Henry!' I yell and, before clicking the escape button, dissolve into a stream of four-letter words.
The fact that my phone's speech-recognition capability struggles to catch a single word raises the question of how any program could precisely transcribe whole sentences onto a screen.
Experience strengthens my doubts. The last time I tangled with speech rec, five years ago, it failed to live up to the hype. But it may deserve a second chance because in theory it means you can sit back and relax your shoulders. No more repetitive strain disorder. On paper, it could be a true killer application, catching on like Facebook.
So this week, I don the fitness instructor-style headphones and investigate whether 'talk-into-text' technology is ready for action. The area is a cinch to research because it's far from crowded now that IBM's ViaVoice has been licensed to Nuance, which is responsible for Dragon NaturallySpeaking: one of just two serious contenders.
Dragon NaturallySpeaking by Nuance Communications (Windows XP or Vista)
Because reading and rendering the human voice is such a complex activity, speech rec programs need to be big. Dragon is huge - to accommodate it, you must have a gigabyte of free hard disk space.