[music-dsp] Re: Voice analysis

[jbv]

Hello,

I'm on the music-dsp list and got your recent message (quoted below).
I'm also on the synth-DIY list, and last week someone initiated a
discussion about the possibility to produce vocal sounds with a bank
of analog BP filters, saying that 3 formants were enough.
I answered that knowing the Fc of each formant wasn't enough :
bw and Amp for each formant were also crucial information.
Unfortunately, I didn't have any formant table values at hand to
illustrate my answer.

So my question is : can I forward your formant table values to the
synth-DIY list ?

Thanks.

JB




> "Sarah Thompson" <sarah at telergy.com> wrote:
> >
> > 1. I seem to remember seeing a list of formant frequencies for typical male
> > and female vowel sounds somewhere. Does anyone have a link to this by any
> > chance? Maybe I even saw it posted here.
>
> You did. This may or may not be it:
>
>   2300|ee
>       |        i
>       |                e
>       |                      aa
>       |            er         ar
>       |         u          a
>    800|  uu          aw
>       |__________________________
>      250                       800
>
>     Heed   Hid  Head   Had   Hod Hawed  Hood Who'd   Hud Heard
> f1   270   390   530   660   730   570   440   300   640   490
> f2  2290  1990  1840  1720  1090   840  1020   870  1190  1350
> f3  3010  2550  2480  2410  2440  2410  2240  2240  2390  1690
>
> And in Hungarian...
>
>         F1      F2       F3       B1     B2     B3     B4     AM
> A':    698     1337    2400     125     125    50     125     14
> A:     554      988    2400     50       50    50     50      11
> O:     391      740    2400     50       50    50     50      8
> U:     267      622    2400     125      50    50     50      5
> U::    233      1897   2400     50       50    50     50      10
> I:     202      2214   2842     125      125   50     309     11
> E':    368      2047   2842     50       50    309    726     13
> O::    415      1528   2400     50       50    50     50      12
> E:     523      1897   2400     125      125   309    726     13
>
>
> > I know that cepstral transforms are used in voice analysis for speech
> > recognition, but I'm not convinced that approach would give the right kind
> > of numbers for what I'd like to achieve.
>
> Cepstrum is probably ok, as is LPC analysis, but you still have to pick out
> the formants yourself. Another method is just to pre-emphasize a smoothed
> (or low resolution) FFT and notch out the highest peak and call it a formant
> then look for the next highest peak and so on...
>
> Look for the software called "Praat" which does various types of speech
> analysis.
>
> Paul.
>
> -----------------------------
>   m a x i m | digital audio
>  http://www.maxim.abel.co.uk
> paul.kellett at maxim.abel.co.uk
>
> dupswapdrop -- the music-dsp mailing list and website: subscription info,
> FAQ, source code archive, list archive, book reviews, dsp links
> http://shoko.calarts.edu/musicdsp/





dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/