where to find formant tables and resonance models?
we probably all know the 25 years old ircam vowels table, but i am in need of more and different ones. does anyone know of a data set which is different or eventually also more complete than the ircam one?
for example i am currently using the ircam tenor data for the systhesis of a falsett voice - but i dont really like the result. i would also like to have different sets for "male", "female" and "children". the tenor and mezzosopran range, for example, applies in real life for both genders.
the ircam data is probably already made from averaging a bunch of different recordings. but the more data i had, the more it would be possible to create an artificial ruleset of how different voices sound like and how the relate to each other (so that you can describe transitions between them)
if you know what a colorspace is - i am looking to create a vocal formant space...
the same request i had for resonance models. it seems impossible to find some onb the net.
what i would like to have is something like 10 or so different acoustic guitar body resonance models and stuff. i know that i could start using impulse-response sample collections of such recordings and process resonance models from there, but this is a lot of work and the time value will still be a matter of trial and error (with no reference recording to check against)
btw, what about consonants? :) ssss-shhhh-hhhh ?
in an hypothetic, ideal case, the "vocal format space" would be multidimesional:
- a. aa, ae, e, i, ..... ö, ü
- tessiturae, from black to falsett
- a set of e.g. 4 different timbres
- chest, head, throat
- male, female, child
- maybe even registers? (those could be created artificially maybe)
i almost said "black, white" but this is probably bullshit from a technical standpoint.
ok i guess we can file this under "question too complicated".
I guess so... though I'd be interested to know more about this.
The only lead I have is from the book Designing Sound by Andy Farnell - here he builds a simple vocal tract model in PD using a list of vowel formants. I think I convereted it to Max at some stage and posted it here, though I now can't seem to find it.
I'm sure I could dig it out if you're interested, but the data was limited (similar to this - if that's the Ircam one you refer to). However, he references The spectral database at Center for Spoken Language Understanding, Oregon University - compiled by Tim Carmell. Perhaps a lead?
i´ve seen this before but didnt use it. it adds 3 additional vocals to what ircam offers - but otoh it lacks gain coefficients.
of course you can use your ears and match these 2 databases together somehow ... :)
Heads up for necroposting!
Not for exactly spoken sounds, but there’s a Diphone Studio in the IRCAM forum software set that can create resonant models from arbitrary sounds. I used them with the CNMAT’s [resonators~] object.
Jumping in years later while doing some grad research on formants for Max/PD programming (we're using the Farnell book as well).
Not sure if this is what was requested but I found these pages with good info & some formant plots over vocal ranges that seemed decent and may give more data options to work with:
https://www.classes.cs.uchicago.edu/archive/1999/spring/CS295/Computing_Resources/Csound/CsManual3.48b1.HTML/Appendices/table3.html
https://home.cc.umanitoba.ca/~krussll/phonetics/acoustic/formants.html
http://ec-concord.ied.edu.hk/phonetics_and_phonology/wordpress/learning_website/chapter_2_vowels_new.htm
Another little bump.
I found this funny/simple/rather realistic/very illustrative voice generator called Pink trombone. It's implemented in javascript and if you reveal the page' source code you can see the whole stuff, well organized, with original scientific publication sources. And it's kinda CC-BY licensed.
i ended up mixing the two existing data tables back in the days by freely (i.e. by ear) interpolating the data.
of course for the three given applications i drive with it (biquad, reson, and paf) different settings are needed in order to get a nice result. so now i have a "formant lab" which only uses the table data as start preset, then lets me create new filters.
however, human voice is interesting, but not all.
ready-made tables for musical instruments would also be great to have.
(my gamma version of diphone for OS9 is somehow out of order)
Thanks TFL. Exactly what I was looking for.
Found an implementation on github by Yonatan Rozin :