2

According to “Google iPhone voice-recognition tool baffled by British accents”, mobile phones have trouble recognizing speech.

Which accent (probably US) works satisfactorily? Are there any resources like videos that demonstrate an accent that works well?

5
  • 1
    It seems to have been designed in California, where the English spoken is essentially General American with the cot-caught merger. So presumably this accent is the best one to use. I'd like to know whether it can recognize a strong Southern accent, as well. May 30, 2011 at 14:18
  • 1
    @Peter Shor Thanks a lot for your links! Since I changed the language details (android) I no longer had issues. I'm really happy that my question wasn't closed as off topic before I got feedback
    – stacker
    May 30, 2011 at 16:12
  • @Peter, that explains it. It only has to be trained to recognise "it's like" and "you know" and it can do Californian.
    – mgb
    May 30, 2011 at 18:14
  • @Martin - as a native Californian, I am offended by your callous stereotyping. You left out "totally" and "fer sure!"
    – MT_Head
    May 31, 2011 at 1:20
  • I use Dragon NaturallySpeaking in English with a French accent, the speech recognition accuracy is high. Dec 20, 2016 at 3:37

1 Answer 1

3

It's not that any particular accent is "more satisfactory" per se, but rather that a system will work better on accents broadly close to those that it has been trained on.

Speech recognition is essentially a statistical process: the system pulls in various acoustic cues and then tries to predict which word/sequence of words is the statistically most likely, given its "knowledge" of which acoustic cues in which combinations tend to indicate which speech sounds. It's difficult to say in advance for a particular utterance which feature(s) will be more prevalent in that prediction: it depends on the particular utterance and on the training data. So it's difficult to point to a specific feature of a particular accent in isolation that will be more or less satisfactory: the system is really much more complex than that. (I disagree from that point of view with Peter Shor's comment above: I don't think you can be so specific as to point to one particular feature and then say "that's the best accent" overall.)

If you're interested more in the subject, then I'd recommend introductions to speech recognition such as you'll find in:

  • Jurafsky & Martin, "Speech and Language Processing"
  • The Oxford Handbook of Computational Linguistics
  • Russel & Norvig, "Artificial Intelligence: A Modern Approch"

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.