1
$\begingroup$

I am currently trying to understand why LPCC and LPC is being used in speech recognition and why?

LPC as I understand is model similar to source filter model, modelling speech production, and LPC thereby tries to estimate the given speech signal, with minimal error.

So we having an estimated signal of the original… but how is that useful in speech recognition task?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

I think a high-level explanation is the right thing to give here:

For recognition of a producing process (in this case: speech) from a signal (in this case: audio), you need to find a mapping from signal to a set of features.

The closer (correlated) these features are to the parameters of the producing process (here: voicing utterances), the more information your feature set contains about the thing you want to recognize; the better your recognition can work.

And since linear predictors have proven very effective in reducing the amount of data needed to transport a voice audio signal (for audio compression!) whilst keeping understandable, it is intuitive that using this thus demonstratably well content-correlated (at least to the human listener) set of coefficients for speech recognition.

It's all about reducing your signal to less data whilst keeping the maximimum of info about what you want to recognize and simultaneously dropping as much info about what doesn't contribute to the phenomenon of interest!

Improve this answer
$\endgroup$
8
  • $\begingroup$ Thanks for the High-level approach.. So LPC basically removes "noise" or a data compression, specified to speech, so sort of an preprocessing step? I sort of still feel like I am missing a link in between in what the actual feature is, and how it is generated.. $\endgroup$
    – Bob Burt
    Commented Oct 3, 2017 at 13:37
  • $\begingroup$ If you want so, yes. It reduces the audio samples (==time coefficients) to a set of coefficients that seemingly describe what the speaker wanted to say pretty well. Then, you can use whatever machine learning algorithm you want to map these far fewer coefficients to actual words $\endgroup$
    – Marcus Müller
    Commented Oct 3, 2017 at 13:39
  • $\begingroup$ hmm... but LPCC in it itself are used for speech recognition.. I understand the hype with MFCC, as the feature is filtered based on how the ear hear sound, but how does the altering the speech, to what ideally is thought to be how speech is produced help here?... What kind of features are extracted.. $\endgroup$
    – Bob Burt
    Commented Oct 3, 2017 at 13:49
  • $\begingroup$ That doesn't matter – what matters is that these feature vectors correlate well with what was said. $\endgroup$
    – Marcus Müller
    Commented Oct 3, 2017 at 13:50
  • $\begingroup$ So the features are basically the "filter components" of how glottal, larynx and so on is designed.. But aren't they static?, and would LPCC thereby only work for single speaker systems.. $\endgroup$
    – Bob Burt
    Commented Oct 3, 2017 at 13:53

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.