Harmonic Product Spectrum

Hi

Can someone elaborate on what Harmonic Product Spectrum is all about.
How is it used to find the fundamental frequency of a music signal?
Im a newbie.
Thanks
Aishwarya

On Dec 20, 9:48 pm, cyberaishu <cyberai...@gmail.com> wrote:
> Can someone elaborate on what Harmonic Product Spectrum is all about.
> How is it used to find the fundamental frequency of a music signal?

Many musical instruments produce notes containing energy at
multiple harmonically related frequencies.  If you scale down
a spectra in frequency by integer multiples equal to the
expected harmonics, then, in an overlay of the several
compressed spectra the harmonics of a note will overlap.
Non-periodic noise is far less likely to have these overlaps,
and an overtone of a note will usually have a lower number of
overlaps than the fundamental frequency.  So one can often
just pick the point in the unscaled spectra with the maximum
amount of overlapping spectral peaks with the frequency
scaled spectra.

You have to be careful if you just take the product of all
the scaled spectra, as any missing fundamental or overtone
will produce a product of zero; so you may have to floor
your spectrum above zero.  And if you take the log of the
spectra before doing a harmonic product spectrum, you end up
with something related to a cepstrum, which is also used for
fundamental frequency determination.



IMHO. YMMV.
--
rhn A.T nicholson d.0.t C-o-M
 http://www.nicholson.com/rhn/dsp.html

On Dec 21, 12:29 pm, "Ron N." <rhnlo...@yahoo.com> wrote:
> On Dec 20, 9:48 pm, cyberaishu <cyberai...@gmail.com> wrote:
>
> > Can someone elaborate on what Harmonic Product Spectrum is all about.
> > How is it used to find the fundamental frequency of a music signal?
>
> Many musical instruments produce notes containing energy at
> multiple harmonically related frequencies.  If you scale down
> a spectra in frequency by integer multiples equal to the
> expected harmonics, then, in an overlay of the several
> compressed spectra the harmonics of a note will overlap.
> Non-periodic noise is far less likely to have these overlaps,
> and an overtone of a note will usually have a lower number of
> overlaps than the fundamental frequency.  So one can often
> just pick the point in the unscaled spectra with the maximum
> amount of overlapping spectral peaks with the frequency
> scaled spectra.
>
> You have to be careful if you just take the product of all
> the scaled spectra, as any missing fundamental or overtone
> will produce a product of zero; so you may have to floor
> your spectrum above zero.  And if you take the log of the
> spectra before doing a harmonic product spectrum, you end up
> with something related to a cepstrum, which is also used for
> fundamental frequency determination.
>
> IMHO. YMMV.
> --
> rhn A.T nicholson d.0.t C-o-M
>  http://www.nicholson.com/rhn/dsp.html

Hi
I ve windowed the signal and taken FFt.
  1.Why shld i downsample it?
  2.What exactly should i do after downsampling?

Please elaborate.
Cheers
.