Frequency extraction from digitized monophonic musical sound (Help)

Following upon Dr. Can Akkoc's post regarding frequency
algorithm (unavailable in MIDI):

Hello -

In the early 1990s, I developed a C-program (with DSP
acceleration enhancement) which produced a time trajectory
of frequency for a recorded musical instrument sound or
voice sung note. The method was developed from software
developed by Lippold Haken and Kurt Hebel at the Univ. of
Illinois CERL group.

In my paper: "The Cro-Magnon Additive Analysis/Synthesis
System" which I gave at the 1991 ICMC (International Computer
Music Conference) meeting held in Montreal, I wrote (regarding
frequency extraction):

"At the heart of this process is an autocorrelation algorithm
which determines the per-sample pair average absolute value of the
numerical difference between corresponding samples in two adjacent
trial windows which meet at the time point for which f0 (the
fundamental frequency) is being estimated (Czvitkovicz 1984 - A
Digital Pitch Extractor - Univ. of Illinois MS Thesis, Dept. of
Electrical and Computer Engineering). As the trial window lengths,
expressed in number of samples per window, are varied over a range
containing the fundamental period length (corresponding to f0), a
clear cut V-shaped minimum in this average difference value occurs
at the fundamental period length (possibly non-integral) over the
most of a pitched sound. Over regions of the sound near the attack
onset and at its very end, f0 must be estimated by the user from less
regular average difference vs. trial f0 screen plots."

I tested the sound analysis software I'd developed by analyzing
sounds with frequency extraction followed by additive analysis of
the sound into a series of time amplitude and frequency envelopes for
the sound's individual partials - up to about 15 kHz - which would
be about the 60th partial for a note with fundamental around middle
C (260 Hz or so). I then digitally resynthesized a sound based on
the set of envelopes developed through the analysis. In most cases,
the sounds were nearly or completely indistinguishable from direct
recordings of the original sounds. The resyntheses often could, upon
close listening, be picked out because their onsets differed slightly
from those of the original recordings.

This testing confirmed to me that the analyses were fundamentally
sound, although they could be improved in small ways.

I'm not aware of work in this area as detailed as mine which has
been done recently, but I have not kept up with the field. I have been
nearly incapacitated for the past few years by progressive undiagnosed
iron deficiency anemia which had so weakened me that even doing the
simplest activities required great struggle and mental and physical
effort. I have been under treatment for this and am experiencing an
improvement - restoration - of mental and physical energy which I had
not believed would ever be possible! I have not yet completely
recovered, however.

The work I had been doing on sound analysis and synthesis hasn't
been completed and it isn't in a neatly packaged form. However, I would
like to make it available to anyone who would be in a position to
make use of it to serve their purposes. There is the possibility that
over the course of the next year or several years, I will be able to
pursue this work to a "closure" where it is in a more generally
usable form. It may be that this work has already been done somewhere
else, and if that is the case, I would not want to put a lot of effort
into duplicating work which has already been done. I'll need to research
this.

Best wishes to all for a happy and successful 2003.

Dave Hill Borrego Springs, CA

--- In tuning@yahoogroups.com, "alternativetuning
<alternativetuning@y...>" <alternativetuning@y...> wrote:
> --- In tuning@yahoogroups.com, ASCEND11@A... wrote:
> > I tested the sound analysis software I'd developed by
analyzing
> > sounds with frequency extraction followed by additive analysis of
> > the sound into a series of time amplitude and frequency envelopes
> for
> > the sound's individual partials - up to about 15 kHz - which
would
> > be about the 60th partial for a note with fundamental around
middle
> > C (260 Hz or so). I then digitally resynthesized a sound based
on
> > the set of envelopes developed through the analysis. In most
> cases,
> > the sounds were nearly or completely indistinguishable from
direct
> > recordings of the original sounds. The resyntheses often could,
> upon
> > close listening, be picked out because their onsets differed
> slightly
> > from those of the original recordings.
> >
>
> Did you ever test sounds from wind instruments, singing voices, or
> bowing strings? At the SpecMus list we are having some discussion
of
> whether or not the partial tones for these sounds are precise
integer
> multiples or not.
>
> Gabor Bernath

a rather definitive answer for this question came up, which mentions
a new, very accurate frequency-extraction technique:

The Journal of the Acoustical Society of America -- February 1996 --
Volume
99, Issue 2, pp. 1210-1218

Frequency ratios of spectral components of musical sounds
Judith C. Brown
Physics Department, Wellesley College, Wellesley, Massachusetts 01281
and
Media Lab, Massachusetts Institute of Technology, Cambridge,
Massachusetts
02139

Abstract:
A recently developed high resolution frequency tracker [J. C. Brown
and M.
S. Puckette, ``A high resolution fundamental frequency determination
based
on phase changes of the fourier transform,'' J. Acoust. Soc. Am. 94,
662-667
(1993)] has made it possible to measure the ratios of the frequencies
of the
upper harmonics of a sound with respect to its fundamental frequency
with
high accuracy. Calculations were carried out on digitized sounds
produced by
a clarinet, alto flute, voice, piano, violin, viola, and cello. The
sounds
produced by the stringed instruments included examples played
pizzicato and
bowed both with and without vibrato. Measured ratios were exactly
equal to
integers for all instruments except the piano and strings played
pizzicato.
Anomalous behavior was observed for the fundamental frequency for
vibrato
sounds played by stringed instruments with the frequency deviation
exceeding
the extrema of the other harmonics divided by their harmonic number
by about
1% on average. Piano inharmonicity was proportional to harmonic number
squared in agreement with Fletcher [J. Acoust. Soc. Am. 36, 203-209
(1964)].
The major limitation on this calculation was found to be instrumental
fluctuations.

it would seem that dave hill or anyone else interested should study
this Brown and Puckette article of 1993.