Almost perfect: 98% accuracy for Fonix speech recognizer

Image for the article Almost perfect: 98% accuracy for Fonix speech recognizer

It’s still not perfect, but it’s getting close to. Fonix claims the last release of VoiceIn could now recognize up to 98% of human voice sounds. The speech-analysis software runs on Linux platform, with supports for Microsoft Windows and Apple Mac. And targets PDAs and mobile phones, medical devices, and multimedia players. VoiceIn is already used in the XBox gaming console, and we guess it will land very soon on the Greenphone, the Trolltech 100% open source smartphone, that might become a touchless handheld.

Oct 21, 2006 | By Nuno

Tags: , , , , , , , ,

3 comments

  • #0 Houlton:
  • 98% accuracy means 2% error. Or one charachter in 50. Assume a document’s average characters per word is 6. That means 50 characters spans about 8.33 words. There are probably close to 100 charachters on a typical document’s line of text.

    Thus at 98% accuracy we are talking about one error every 8.3 words, give or take. (Yes, I know the arithmetic is not quite that simple.) And about two errors every text line.

    Not good enough. Soon perhaps, but not yet.

  • #1 hombrelobo:
  • I agree, 98% is bad. I remember IBMs system, ViaVoice I think the name was ( :) ), having a 96% so …. 2% is a great improvement in 5 years ???

    Unless we get 99.99% the systems are useless. Can you imagine a keyboard with 98% accuracy ???

  • #2 imski:
  • It’s not going to make errors with letters in words though is it?

    The 2% error is only likely to affect whole words or discreet letters like a, and I.

21talksTracking the telecoms evolution