I am trying to generate vowels on MATLAB by using source-filter model. In case of monophthongs I'm using Rosenberg pulse as my source signal and then this source signal is convoluted with sounds generated from formants. Basically I'm using the approach used in the following project http://in.mathworks.com/matlabcentral/fileexchange/45449-vowel-synthesis

Now I'm trying to do synthesis of diphthongs which are basically generated by gliding from one monophthong to another. I've tried weighted-average but it's not working. So how can I model this transitioning from formants of one monophthong to other to generate the IPA diphthongs ?

  • I do remixes and sometimes I need "faking" someone's words from existing speeches. This includes "gluing" vowel phones (V) into diphthongs (VV). I found that gliding fails (sounds harsh) if the vowel tones are of too different pitch. This is not am issue in VC (vowel+consonant) or CV patterns. Try making your phone corpus match a single pitch. Also, I'm not familiar how the tool you're using mixes phones, but make sure that VV pattern always "cross-fades" two phones. Unlike VC/CV patterns, where a simple addition works fine. Commented Apr 13, 2016 at 10:27
  • Also, search for questions at Sound Design.SE as they have some topics for sound synthesis and MATLAB. Commented Apr 13, 2016 at 10:38
  • 1
    @bytebuster cross-fading works and diphthongs can be produced sufficiently good. Thanks a lot. Commented Apr 14, 2016 at 8:58


Your Answer

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