I want to know that what features might be needed for recognizing a word i a sentence exactly? I guess the routine for recognizing the word in a sentence is ganna follow this: 1-calculate voiced and unvoiced and silence frames 2-determine a threshold 3- when ever we get to a silence frame,that could be an end point of a word besides considering other features ... This is the main idea i guess , and if it is wrong or right or anything else,i would really appreciate if you help me out with completing these steps... Thanks

  • 2
    $\begingroup$ In normal speech there are no gaps between words - the words are elided, so separating them is quite a challenge. This is why early speech-to-text software forced you to speak each word separately. $\endgroup$ – Paul R Feb 6 '14 at 11:54
  • $\begingroup$ @PaulR so your telling me that first of all these, we have to transform speech to text? Is there any special app you introduce me to test? thanks $\endgroup$ – drizzle Feb 6 '14 at 12:58
  • 1
    $\begingroup$ No - I was just giving some context - segmenting natural speech into words is a non-trivial problem. $\endgroup$ – Paul R Feb 6 '14 at 14:21

In sentences in normal human speech, words are usually not separated by silent frames or thresholds. Humans guess (perhaps some form of expectation driven pattern matching) at word separations, and are sometimes wrong.

| improve this answer | |
  • $\begingroup$ Thanks for your response,but i am really curious that how it can be recognize in machines,in human we dont need any pattern to recognize... so there would be something else we should focus on... but idk what is that! $\endgroup$ – drizzle Feb 9 '14 at 21:01
  • $\begingroup$ Experiments show you can completely leave out a word in a long fast sensible sentence or paragraph (replace the word with background noise of the right volume), and most humans will think they actually heard the missing word. So it is the pattern, because the word isn't even there. Continuous speech recognition software uses a database containing a massive amount of statistical patterns to search. $\endgroup$ – hotpaw2 Feb 9 '14 at 21:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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