Loading ...
Sorry, an error occurred while loading the content.

Formants Extraction Algorithm

Expand Messages
  • Shaikh Faisal Rasheed
    Hi dear I am student of University of Engineering and Technology Lahore Pakistan. I am currently working over the Arabic Speech Recognition System. I use Praat
    Message 1 of 4 , Jun 2, 2003
    View Source
    • 0 Attachment
      Hi dear

      I am student of University of Engineering and
      Technology Lahore Pakistan.

      I am currently working over the Arabic Speech
      Recognition System. I use Praat as the tool for
      feature extraction and use Formants for analysis.
      Since I have gain very fruit full results using
      formants in speech recognition and now I want to
      implement the formant feature extraction in my
      software. In this regard I need some help about the
      Algorithm implemented in praat specially formant
      (burg) so that I implement It. If some one has source
      code for formant burg kindly send it to me.

      Thanks

      Regards

      Faisal


      __________________________________
      Do you Yahoo!?
      Yahoo! Calendar - Free online calendar with sync to Outlook(TM).
      http://calendar.yahoo.com
    • Paul Boersma
      ... until two weeks ago, the published Praat source code did not contain the inner algorithms for formant extraction, since we used the routines memcof and
      Message 2 of 4 , Jun 2, 2003
      View Source
      • 0 Attachment
        At 00:56 -0700 02-06-2003, Shaikh Faisal Rasheed wrote:
        >If some one has source
        >code for formant burg kindly send it to me.

        until two weeks ago, the published Praat source code did not contain the inner algorithms for formant extraction, since we used the routines memcof and zroots from Numerical Recipes, which has a non-open copyright. In the current version, downloadable from the Praat web site, we use code reimplemented from Burg's description, and root finding algorithms from LAPack and comparable open source code, and all this can be inspected when you download it. Remarkably, the formant results for a couple of test sounds were exactly the same as before, up to and including the last 'bit'...
        --

        Paul Boersma
        Institute of Phonetic Sciences, University of Amsterdam
        Herengracht 338, 1016CG Amsterdam, The Netherlands
        http://www.fon.hum.uva.nl/paul/
        phone +31-20-5252385
      • Faisal
        Thanks for your collaboration I downloaded the code of praat. The code is for Linux and have make file for linux operating system. This file is not supported
        Message 3 of 4 , Jun 6, 2003
        View Source
        • 0 Attachment
          Thanks for your collaboration I downloaded the code of
          praat.
          The code is for Linux and have make file for linux
          operating system. This file is not supported by the
          Visual C namake command and directly supported make
          file feature.
          Actually I want to compile this code in VC. I only
          required to extract formants from the input speech of
          a particular phoneme then pass it to the recognition
          engine, which I have, build. I will be very grateful
          to you if you help me in this regard.

          I studied the help in praat and saw the algorithm for
          formant to burg but this not help me fully.

          The steps are
          1) Resample the sound according to the highest formant
          value. How?? I cant understand the Algo written in the
          praat help
          2) Pre-emphasize the sound using the gaussian like
          window with 50 hz. But what should be the Delta t
          (sampling period)?
          3) Then Compute the LPC coefficients.

          Now how to compute the Formants after computing the
          above three steps?

          Actually I need the frequency values in each frame
          of each formant which praat calculate when we apply
          the command on sound file to To Formant (burg)�


          If some one take time and send me the Algo, that how
          to use the source code of praat to extract the
          formants only or give me some hints I am very
          thankful.

          Thanks
          Regards
          Faisal



          --- Paul Boersma <paul.boersma@...> wrote:
          > At 00:56 -0700 02-06-2003, Shaikh Faisal Rasheed
          > wrote:
          > >If some one has source
          > >code for formant burg kindly send it to me.
          >
          > until two weeks ago, the published Praat source code
          > did not contain the inner algorithms for formant
          > extraction, since we used the routines memcof and
          > zroots from Numerical Recipes, which has a non-open
          > copyright. In the current version, downloadable from
          > the Praat web site, we use code reimplemented from
          > Burg's description, and root finding algorithms from
          > LAPack and comparable open source code, and all this
          > can be inspected when you download it. Remarkably,
          > the formant results for a couple of test sounds were
          > exactly the same as before, up to and including the
          > last 'bit'...
          > --
          >
          > Paul Boersma
          > Institute of Phonetic Sciences, University of
          > Amsterdam
          > Herengracht 338, 1016CG Amsterdam, The Netherlands
          > http://www.fon.hum.uva.nl/paul/
          > phone +31-20-5252385


          __________________________________
          Do you Yahoo!?
          Yahoo! Calendar - Free online calendar with sync to Outlook(TM).
          http://calendar.yahoo.com
        • Paul Boersma
          ... you can just import the source code into Visual C++. It is not just for Linux. With the same code, we create the Macintosh, Windows, SGI, Solaris, and HPUX
          Message 4 of 4 , Jun 11, 2003
          View Source
          • 0 Attachment
            At 7:04 -0700 6-6-2003, Faisal wrote:
            >The code is for Linux and have make file for linux
            >operating system. This file is not supported by the
            >Visual C namake command and directly supported make
            >file feature.

            you can just import the source code into Visual C++. It is not just for Linux. With the same code, we create the Macintosh, Windows, SGI, Solaris, and HPUX editions.

            >1) Resample the sound according to the highest formant
            >value. How?? I cant understand the Algo written in the
            >praat help

            A sampled sound can be regarded as a summation of sin(x)/x functions (see books on signal analysis). Praat's interpolation computes points along this reconstructed curve.

            >2) Pre-emphasize the sound using the gaussian like
            >window with 50 hz. But what should be the Delta t
            >(sampling period)?

            the sampling period is the inverse of the sample rate. If the sample rate is 22050 Hz, the sampling period is 1/22050 seconds.

            >3) Then Compute the LPC coefficients.
            >
            >Now how to compute the Formants after computing the
            >above three steps?

            the last step is to find the zeroes of a polynomial. See books on speech analysis and the source code file fon/Sound_to_Formant.c.

            --

            Paul Boersma
            Institute of Phonetic Sciences, University of Amsterdam
            Herengracht 338, 1016CG Amsterdam, The Netherlands
            http://www.fon.hum.uva.nl/paul/
            phone +31-20-5252385
          Your message has been successfully submitted and would be delivered to recipients shortly.