February 06, 2008

Any oxymoron up there with "Military Intelligence" and "Microsoft Works"...

Writing about web page http://www.bbc.co.uk/consumer/tv_and_radio/watchdog/reports/consumer_goods/consumer_20080204.shtml

On Monday, Watchdog did a report on the DS and how Brain Training doesn't recognise some people's voices. As much as I like Watchdog, there are two reasons why this report was stupid, pointless, and irrational:

  1. All voice recognition is crap; they all have trouble with accents and general recognition.
  2. Even people have trouble with accents (I refer you to complaints about Indian call centres that do not centre around job availability).

Here's a video to back me up:

- 2 comments by 0 or more people Not publicly viewable

  1. Timothy

    Yes. Speech recognition has always had problems. The problem with Microsoft was not that they didn’t manage it but that they attempted to prove that they had when they hadn’t. The “Ambient noise” excuse is no excuse at all. The FIRST thing I’d want a speech recognition device to do is to find some way (hardware or software or for best effect, combined) to eliminate or differentiate from background noise.

    Back in the early days of speech recognition Apple had a much more humble approach. Upon completing the first version of their speech recognition software Apple issued all the programmers who worked on the project with a t-shirt bearing the legend “I helped Apple to wreck a nice beach”. (If you don’t get it straight away try saying it aloud – you’ll be laughing within the next five minutes).

    17 Apr 2008, 14:35

  2. Timothy

    Sorry. Didn’t read what you said clearly and I’m a bit of a video magpie so I clicked on the youtube without reading properly.

    I wonder whether the DS worked well with regional Japanese accents. I mean, if you cracked it with one language you’d feel confident about launching it in another country – with disastrous results (you’d have to do the legwork all over again for each country due to the massive differences in accent and dialect in each country – I can understand a dutchman speaking English a lot easier than I understand someone who comes from Newcastle even though they come from the UK like I do). Accent and language recognition is key to voice recognition. That and the elimination of extraneous noise. Perhaps they should work on first getting a device/program that could accurately write down what is being said using the international phonic alphabet. Once that has been done computers and humans could more effectively analyse speech patterns. Perhaps a camera should be used alongside a microphone in order to provide extra visual clues as to which phonemes are being formed. There’s an awful lot of subtlety though. In one Australian aboriginal dialect the difference between two ‘n’ sounds is that one is pronounced by curling the tongue back and touching the roof of the mouth with the tip and the other is pronounced the same as the European ‘n’ by squashing the tongue flat up against the roof of the mouth. I’ve tried both and in isolation they sound exactly the same to me. They probably affect the other sounds produced before and afer though. It will be an age before we crack this. Maybe some of the ‘silicon brain’ solutions that have produced the digital retina and the enhanced cochlear implant will be helpful in cracking this problem. For the moment it’s still a technology of the future.

    17 Apr 2008, 14:58

Add a comment

You are not allowed to comment on this entry as it has restricted commenting permissions.

February 2008

Mo Tu We Th Fr Sa Su
Jan |  Today  | Mar
            1 2 3
4 5 6 7 8 9 10
11 12 13 14 15 16 17
18 19 20 21 22 23 24
25 26 27 28 29      

Search this blog


Google PageRank


Not signed in
Sign in

Powered by BlogBuilder