Vocalist.org archive


<
From:  "Kevin Hollis" <khollis@m...>
"Kevin Hollis" <khollis@m...>
Date:  Thu Nov 22, 2001  12:54 pm
Subject:  Re: [vocalist] Know of voice research using spectral analysis?


> What I'd like to do is record a bunch of mezzos, sopranos, countertenors
> and tenors and compare the position/formation/intensity of the frequency
> bands, especially the singers formant. I will use spectral analysis to
> quantify differences between the voices, if any.
>
> So... all you vocal teachers and scientists out there... I was hoping you
> would critique my research proposal and suggest related research papers.
> I'd love to see what's been done so far in this vein!
(spectal/layrgiscopic
> studies comparing voice types) Also, if you can offer any words of advice
> on how to make this project more effective or easier, i'd appreciate it!

Web search fo 'Ingo Titze'.
Martyn Clark may be able to help or point you at an appropriate website, dig
back through Vocalist for old mails of his.

(Scientific opinion from a physicist not based in vocal research, but open
to offers if someone wants to offer me a job ;-)
I'd reduce your variables as much as possible, rather hard when you have the
infinite variability of humanity. How do you compensate for 'singer on a bad
day', 'bad technique giving poor vowel formants' etc... etc... ? In order to
get a good scatter graph for each voice category you would want at least 30
singers of each 'type'. Of course, type is really splitting a continuum into
sections, which is a poor measure. Maybe plot against 'highest comfortable
note' rather than voice category, males as red dots women as green.
Representing the data you obtain will require as much thought as planning
the experimental details. Search for 'nomogram/nomograph'; these allow many
variables to be plotted against a single x-axis and look at relationships
between variables. Also there are several good 3-D plotting packages around
(Origin was my choice 5 (gulp) 7 years ago, things have probably moved on
though.) You can get to 4-D using colour and 5-D using intensity of colour.
Really you would need to understand how 'formants etc...' vary for a single
voice type with age / sex / height / years of training / state of health /
diet / fluid intake / neck length / larynx diameter / mouth cavity shape
etc.. etc... before trying to compare singers of different vocal types. That
way you could choose singers from different vocal types with comparable
'other' parameters. But then again I guess you haven't got 30 years !!
Gather as much data as you reasonably can from the singers that you can get,
you may find relationships that you hadn't expected and are far more
interesting. This is IMHO where really good science comes from.

I'd forget the Italian song and stick to really simple vocalises with as
many singers as you can muster.

When writing your conclusions make sure that you're not too dogmatic about
what you have found. Unless you have huge resources at your disposal it is
very likely that your data will be 'open to interpretation'. Point out the
problems in the data before your examiner/supervisor does !

kevin


emusic.com