Re: Robust method of fundamental frequency estimation.

[Anssi Klapuri <klap@xxxxxxxxx>]

Dear Roisin,

Based on my experience with the multipitch detection of musical
sounds, I would suggest the following:
- -At low pitch values, autocorrelation- or comb-filter
  related methods are better than spectral methods.
  This is because the F0 resolution of the ACF is much better
  than that of FFT at low frequencies. You may check this by
  calculating the F0 difference between two ACF lags at low-pitch
  lags and the frequency difference of two FFT bins at low frequencies.
  Of course, ACF can be implemented in frequency domain,
  and in principle any frequency-domain method can be implemented
  in the time domain, but this is another story.
  More generally speaking, you have to utilize the entire harmonic
  series of a sound to improve pitch resolution in short frames.
- -Spectral whitening (flattening the spectral energy distribution)
  by inverse-LPC filtering or other methods is essential in order
  to achieve robustness for different instruments.
  This is especially true for low-pitched sounds with strong formants.
- -To reduce the effect of inharmonicity at the lowest strings,
  you may try to lowpass filter the signal so that only
  the lowest 10-20 harmonics remain, but I do not know if this helps.

Among my own publications, you may have a look at:
  Klapuri, A., " Multiple fundamental frequency estimation by summing
  harmonic amplitudes," 7th International Conference on Music Information
  Retrieval (ISMIR-06), Victoria, Canada, Oct. 2006.

The above method works well for sounds down to 40 Hz in pitch, but
below that irregularity is the commonplace and I did not try.
The method is based on implementing a comb-filter in the freq domain.

Best regards,

--Anssi
___________________________________________________________________________
Anssi Klapuri   klap@xxxxxxxxx   http://www.cs.tut.fi/~klap
Institute of Signal Processing, Tampere University of Technology
P.O.Box 553, FI-33101 Tampere, Finland
office: +358 3 3115 2124, gsm: +358 40 849 0784


Date:    Wed, 31 Jan 2007 17:11:52 +0000
From:    Roisin Loughran <roisinaud@xxxxxxxxxxx>
Subject: Robust method of fundamental frequency estimation.

--0-619710193-1170263512=:68894
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: quoted-printable

Dear list,=0A=0AI was wondering if any of you know the most robust way to c=
alculate the fundamental frequency of a note across the range of a variety =
of instruments?=0A=0AI'm currently working on a matlab program and have tri=
ed using the auto-correlation method and the cepstrum method but have found=
 that these both have difficulty in calculating f0 of timbre-rich tones suc=
h as those from a piano - particularly in the lower pitch ranges. Does anyo=
ne know of a method that is more reliable in these regions or is it necessa=
ry that I investigate such complex tones by a different means? From examini=
ng a number of the FFTs from these signals it is tempting to just pick the =
first strongest partial - the complex overtones just seem to confuse the mo=
re complicated algorithms, but I realise that this is hardly a reliable app=
roach.=0A=0AAny suggestion would be greatly appreciated,=0AThanks in advanc=
e,=0A=0ARoisin Loughran=0A=0A=0A=0A=0A=09=09=0A____________________________=
_______________________________ =0ATo help you stay safe and secure
online,=
 we've developed the all new Yahoo! Security Centre.
http://uk.security.yah=
oo.com

--Anssi
___________________________________________________________________________
Anssi Klapuri   klap@xxxxxxxxx   http://www.cs.tut.fi/~klap
Institute of Signal Processing, Tampere University of Technology
P.O.Box 553, FI-33101 Tampere, Finland
office: +358 3 3115 2124, gsm: +358 40 849 0784