2
$\begingroup$

Is it possible to compare two audio samples(meaning find word from one in another) realtime with cross-correlation?

$\endgroup$
1
  • $\begingroup$ No I'm iOS developer and I'm searching the way to detect predefined word in real-time:) $\endgroup$ – user4947 Jul 11 '13 at 8:27
7
$\begingroup$

No. If the words are pronounced with a different speed or intonation (prosody), or by different speakers, the cross-correlation won't have any significance.

A solution more robust to changes in prosody is to use audio features which are less dependent on pitch (such as MFCCs), and a comparison method that is robust to local time stretching (such as DTW or a distance between HMMs). It will still perform poorly when comparing a recording of two different persons saying the same word.

$\endgroup$
2
  • $\begingroup$ thanks for response, i use same device with same input and output. Is there any way to use cross-correlation in this case? $\endgroup$ – user4947 Jul 7 '13 at 20:00
  • 4
    $\begingroup$ No, this is not how things work... People are not record players - every utterance of a word, even by the same person, has differences in timing, pitch, timbre which can alter the shape of the signal. You need to use a signal processing technique robust to these variations and looking at the raw waveform is not one of them. $\endgroup$ – pichenettes Jul 7 '13 at 20:18
3
$\begingroup$

No. Human word recognition is a very complex psycho-perceptual phenomena. Things that sound like the same word to a human can have completely different waveforms with near zero correlation (different pitch frequencies and tonal modulations, timings, formant centers, surrounding context, etc.etc.)

$\endgroup$
0
$\begingroup$

Not possible... what @pichenettes says was right.. "People are not record players - every utterance of a word, even by the same person, has differences in timing, pitch, timbre which can alter the shape of the signal"

unless @user4947 is saying that it is the same audio file but maybe 1 is a cut out version while the other is full audio file..

$\endgroup$

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