WAVE file analysis and FFT

Hi, this is my first post here and was hoping you guys may be able to help
me out :-) I'm currently working on a project at uni to transcribe musical
notes from a recording. At the minute I'm trying to calculate the frequency
of a single note from a wave file. So far using C++ I have read in the
headers etc  and have reached the data chunk. However, I'm not sure what I
need to do with the data to be able to perform an FFT! Can I just read each
bit of the data into an array and do an FFT on that or does the bits per
sample have to be taken into account (eg put 16bits into each array
element if the bps is 16)?

I am currently using the implementation of the fft outline at
http://www.codeproject.com/cpp/howtofft.asp and have had some successes.
Currently analysis a clip of Middle C (freq 261Hz) I get a result of 449hz
but analysis the same note from a recording of my guitar yields crazy
results such as 16000hz (!) and I don't know what could be the issue.

On Sun, 21 Oct 2007 19:20:19 -0500, Irie wrote:

> Hi, this is my first post here and was hoping you guys may be able to help
> me out :-) I'm currently working on a project at uni to transcribe musical
> notes from a recording. At the minute I'm trying to calculate the frequency
> of a single note from a wave file. So far using C++ I have read in the
> headers etc  and have reached the data chunk. However, I'm not sure what I
> need to do with the data to be able to perform an FFT! Can I just read each
> bit of the data into an array and do an FFT on that or does the bits per
> sample have to be taken into account (eg put 16bits into each array
> element if the bps is 16)?

Your array needs to be of properly formed samples, so yes you need to take
the bits/sample into account.  Like good lab practice, good programming
practice involving data acquisition says that you should make a check
plot.  So once you've gotten your data, toss it up on a graph and compare
it to what you see on an oscilloscope clipped onto the microphone leads,
to make sure it looks about right.

Then do your FFT.

> I am currently using the implementation of the fft outline at
> http://www.codeproject.com/cpp/howtofft.asp and have had some successes.
> Currently analysis a clip of Middle C (freq 261Hz) I get a result of
> 449hz but analysis the same note from a recording of my guitar yields
> crazy results such as 16000hz (!) and I don't know what could be the
> issue.

In the first case it sounds like you're not reporting your sample rate
correctly to the FFT routine, or like your data is crap.  In the second
case, it sounds like crap data.

Were I going to do something like this, I'd use Scilab or MATLAB to
prototype my algorithms.  Only when I had them working in mathemagic land
would I start messing around with C or C++.  For mathematically intensive
work like this there's about a 10:1 ratio of effective work that you can
get done with one of the above programs vs. flailing around with doing the
math in C or C++.

-- 
Tim Wescott
Control systems and communications consulting
http://www.wescottdesign.com

Need to learn how to apply control theory in your embedded system?
"Applied Control Theory for Embedded Systems" by Tim Wescott
Elsevier/Newnes, http://www.wescottdesign.com/actfes/actfes.html

Hello,

I have some Matlab example code here, maybe it's useful:
http://www.elisanet.fi/mnentwig/webroot/FFT_peaksearch_audio_example/index.html

-mn

Irie wrote:

> Hi, this is my first post here and was hoping you guys may be able to help
> me out :-) I'm currently working on a project at uni to transcribe musical
> notes from a recording. At the minute I'm trying to calculate the
> frequency of a single note from a wave file. So far using C++ I have read
> in the
> headers etc  and have reached the data chunk. However, I'm not sure what I
> need to do with the data to be able to perform an FFT! Can I just read
> each bit of the data into an array and do an FFT on that or does the bits
> per sample have to be taken into account (eg put 16bits into each array
> element if the bps is 16)?

You have to take care of the bps. I recommend to use libsndfile for reading
audio files, then you don't have to care about these details:

http://www.mega-nerd.com/libsndfile

Tschau
Andreas
-- 
Andreas H�nnebeck | email: acmh@gmx.de
----- privat ---- | www  : http://www.huennebeck-online.de
Fax/Anrufbeantworter: 0721/151-284301
GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc
PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc

On Oct 21, 5:20 pm, "Irie" <headshredder...@yahoo.co.uk> wrote:
> Hi, this is my first post here and was hoping you guys may be able to help
> me out :-) I'm currently working on a project at uni to transcribe musical
> notes from a recording. At the minute I'm trying to calculate the frequency
> of a single note from a wave file. So far using C++ I have read in the
> headers etc  and have reached the data chunk. However, I'm not sure what I
> need to do with the data to be able to perform an FFT! Can I just read each
> bit of the data into an array and do an FFT on that or does the bits per
> sample have to be taken into account (eg put 16bits into each array
> element if the bps is 16)?
>
> I am currently using the implementation of the fft outline athttp://www.codeproject.com/cpp/howtofft.aspand have had some successes.
> Currently analysis a clip of Middle C (freq 261Hz) I get a result of 449hz
> but analysis the same note from a recording of my guitar yields crazy
> results such as 16000hz (!) and I don't know what could be the issue.

Even after you get your file format parsing and FFT
algorithms debugged, there is another bigger issue.

The result from a single FFT and the frequencies for "notes
from a recording" are often neither identically nor even simply
related, even more so for guitar recordings.  You may have
to consider transients, interfering resonances, evolving
harmonic content, string inharmonicity, microphone frequency
response curves, and FFT windowing artifacts, among other
effects.


IMHO. YMMV.
--
rhn A.T nicholson d.0.t C-o-M

Irie wrote:

> Hi, this is my first post here and was hoping you guys may be able to help
> me out :-)

I hope this is your first and last post here.

> I'm currently working on a project at uni to transcribe musical
> notes from a recording.

Forget it it will never work.

VLV

Change your project to another one. It will never work properly. One
can't simply decompose music artwork into FFT. To handle ART one has
to apply similar level math ART. So, it's much better just to play/
learn these music yourself ...

Nordic regards,
Yuri

I could imagine that building a tool to *assist* writing down musical
scores is well within the scope of a project work:

- understand FFT and related things
- demonstrate what can be done
- discuss the limitations

Tracking a monophonic tune is rather straightforward (there is plenty of
"prior art" on the web). A general solution that works with Britney Spears
is a completely different problem.

-mn

PS: If I were to give the task for a study project (not a real thesis!),
my requirements could be as follows:
- FFT-based waterfall display in sync with audio playback
- connect a MIDI keyboard
- when pushing a note on the keyboard, show the fundamental and harmonics
on the waterfall display

Sure, it could be basic tone analyzer project. Use MATLAB to open and
fft audio files first, it reads them directly.

Yuri

On Oct 22, 1:45 pm, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> > I'm currently working on a project at uni to transcribe musical
> > notes from a recording.
>
> Forget it it will never work.

...which is exactly why this might be a good student project
assignment.  One can learn a lot from projects which fail,
or projects that look like they might work given very simple
test cases but not in the general case or in the real world.
This type of project has the additional feature that it might
require a multidisciplinary study of several fields, perhaps
including some basic music theory, perceptual psychology, the
physics of sound, dsp as well as statistical pattern matching
techniques, programming file handlers, and etc.



IMHO. YMMV.
--
rhn A.T nicholson d.0.t C-o-M