[PD] measurement of spectral richness

[Jamie Bullock]

Hi,

On Tue, 2008-02-12 at 13:45 -0800, punchik punchik wrote:
> ive heard a lot of times that natural sounds has more
> spectral richness that digital generated sounds...
> is it possible to measure the spectral richness of a
> sound?

There isn't a standard metric for 'richness' as far as I know. However,
you might be able to break down the 'richness' percept into a number of
features that can be measured. 

Things that spring to mind as contributing to perceived richness are:

- The number of measured partials
- The 'inharmonicity' of the spectrum (i.e. its divergence from a purely
harmonic signal)
- The 'spectral flux' the amount of variation (transience) in the
spectrum between frames

You could try to find a mathematical relationship between these factors
that corresponds to the richness percept, or you could use some kind of
classifier which you would 'train' to relate a given richness input
vector (npartials, inharmonicity, flux) with a perceptually determined
output measure of richness. One of the simplest classifiers for this is
the k-NN classifier. 


Jamie

-- 
www.postlude.co.uk