I am trying to synthesize the French vowels [o] and [ɔ] for running a perception experiment. I have been using the Praat Vocal Toolkit and got pretty nice results with the following formant values: F1(o)=430, F2(o)=800 Hz, F1(ɔ)=490, F2(ɔ)=1000 (I kept the other parameters at their default values). However, the F2 values that I used do not correspond to the typical F2 values one finds in the literature for those vowels. For instance, Gendrot and Adda-Decker (2005) found F2 values of about 1000 Hz and 1200 Hz for [o] and [ɔ], respectively. But, with these values, the vowels sound too central to my ear.

Any idea of why that might be so? And also, does anyone know of another (free) software that I could use to synthesize French vowels with realistic formant values?

  • 3
    You are right, your values seem more reasonable. Your sounds are mid back vowel, so your F2s are logical. Write 'vocoder software' in a search engine and you will get a lot of solutions like audacity. – amegnunsen Jul 5 '19 at 10:47
  • 3
    You may want to check out VocalTractLab. – Nardog Jul 7 '19 at 10:41

I'm not familiar with this toolkit, but given your description my suggestion would be to check formant bandwidths. You are creating rounded vowels, and formant bandwidths are correlated with the degree of lip rounding (see Stevens 2000 p 292). Inaccurate formant bandwidths may lead to perceptual differences even if the formant positions are correct since listeners may rely on it as an acoustic cue to rounding. Rounding also tends to lower F2 values. If the bandwidth cue is missing, lowering F2 may improve perception by providing an additional cue, so this may explain why lowering the F2 values worked for you.

You may find modifying existing recordings more reliable than synthesizing tokens from scratch. By using naturalistic recordings and modifying the formant values, you can retain all the cues listeners attend to but which you may not think to include or manipulate. It seems the toolkit you use has this functionality as Change formants. Will Styer also has a great praat manual which includes a section on source-filter vowel resynthesis (7.15) which allows you to manipulate formant frequencies and bandwidths of existing tokens.

| improve this answer | |

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.