FPGA Central

Hello!

I'm trying to develop an application whose primary purpose is to detect
which tone the user is singing (real time) using frequency analysis, and I
wonder which algorithm to use. I analyze samples of approx 1/10 sec. I first
tried using fourier transform, but since the frequencies are spread out
evenly up to the Nyquist frequency, the resolution is way too bad (approx 3
Hz). I need at least .1Hz, preferrably .01 Hz or better (.03 Hz is one cent
in the vicinity of low A, 55Hz). I then tried the Chirp-Z transform in a
2-pass process where the first pass finds the approximate peak and the
second pass zooms in around that peak to find a more exact value, but it's
considerably more expensive and references to it on the web are very rare,
leading me to believe that its not commonly used. Also, the peak is quite
wide, even for my tuning fork. I do apply a Hanning window before the
analysis.

Can anyone point me in the right direction?

Erik wrote:
> Hello!
>
> I'm trying to develop an application whose primary purpose is to detect
> which tone the user is singing (real time) using frequency analysis, and I
> wonder which algorithm to use. I analyze samples of approx 1/10 sec. I first
> tried using fourier transform, but since the frequencies are spread out
> evenly up to the Nyquist frequency, the resolution is way too bad (approx 3
> Hz). I need at least .1Hz, preferrably .01 Hz or better (.03 Hz is one cent
> in the vicinity of low A, 55Hz). I then tried the Chirp-Z transform in a
> 2-pass process where the first pass finds the approximate peak and the
> second pass zooms in around that peak to find a more exact value, but it's
> considerably more expensive and references to it on the web are very rare,
> leading me to believe that its not commonly used. Also, the peak is quite
> wide, even for my tuning fork. I do apply a Hanning window before the
> analysis.
>
> Can anyone point me in the right direction?

It is a humorous truism among voice instructors that "vibrato should not
exceed the minor third". Is there a singer anywhere who can hold a tone
to within a third of a cent long enough for the actual frequency to be
measured? Even a flautist? I suggest that you think about what you want
to do as well as how to do it.

See if http://www.numerix-dsp.com/zoomfft.html helps. The longer a
signal is sampled, the finer the frequency resolution you get. the
frequency resolution is approximately the reciprocal of the measurement
interval. The signal's appearing to be substantially broader than that
indicates that it has sidebands from amplitude or frequency variations.

Jerry
--
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Hi Erik,

"Erik" <[email protected]> wrote in message
news:Ootxe.28389$[email protected]..
> I'm trying to develop an application whose primary purpose is to detect
> which tone the user is singing (real time) using frequency analysis, and I
> wonder which algorithm to use. [...]

Get "A high resolution fundamental frequency determination based on phase
changes in the Fourier transform", which
describes what I believe is the standard way to accomplish this.

You can get it from Judith Brown's home page here: (there's also a link to
the djvu plugin which you will have to install to see it).

http://www.wellesley.edu/Physics/brown/jbrown.html

--
Matt

Thanx!

If the algorithm is as good as it appears to be in that paper, it's just
what I was looking for.

> Hi Erik,
>
> "Erik" <[email protected]> wrote in message
> news:Ootxe.28389$[email protected]..
>> I'm trying to develop an application whose primary purpose is to detect
>> which tone the user is singing (real time) using frequency analysis, and
>> I wonder which algorithm to use. [...]
>
> Get "A high resolution fundamental frequency determination based on phase
> changes in the Fourier transform", which
> describes what I believe is the standard way to accomplish this.
>
> You can get it from Judith Brown's home page here: (there's also a link
> to
> the djvu plugin which you will have to install to see it).
>
> http://www.wellesley.edu/Physics/brown/jbrown.html
>
> --
> Matt
>
>