@katjav said:

By the way, I do get plenty false pitch reports from [aubiopitch~]. It is a very rough implementation. Hmm. Moog1, could you share your rewrite?

Mmm
If I remember rightly, the threshold wasn't working, amongst other things..

I found my modified aubiopitch~.c .. it compiles .. it still needs a bit more attention, but I found it useful for comparing the different algorithms/methods .. I've attached it below for you to browse / 'play with'

http://www.pdpatchrepo.info/hurleur/aubiopitch~.c

Hi moog1,

Thanks for sharing your rewrite. With the threshold inlet it is now indeed possible to set an appropriate level for the yinfft method. It is not super responsive though. The fcomb and mcomb methods are extremely precise, even for high frequencies. But they do not react to the threshold. Smitt makes Pd crash.

Moog1, did you use elements or aspects of aubio for [basspitch~, or did you instead design your own solution because aubio did not work for your purpose?

Katja

@katjav said:

did you use elements or aspects of aubio for [basspitch~, or did you instead design your own solution because aubio did not work for your purpose?

Katja

I started of by using schmitt ( wasn't crashing for me, but I might have changed something ), due to efficiency/speed, but ended up using fft/freq domain due to its excellent filtering properties .. and as PD already had 'mayerfft' , I got distracted from aubio

Although you are experiencing difficulties, I still think that the aubio libs are good .. could always 'borrow' any code that seems useful/efficient.

I was using libsamplerate to downsample, but decided that a simple interpolation was sufficent/efficient .. I like simplicity, but normally end up compromising to some extent .. simple things are generally more stable, although C++ is reasonably foolproof, there are significant overheads!

@moog1 said:

I still think that the aubio libs are good .. could always 'borrow' any code that seems useful/efficient.

The structure of aubiolib is decent and it's build system as well, it is the work of a professional. It is a pity that the Pd wrappers do not exploit the full power of the routines, but indeed aubiolib is a good source to test various detection methods. It is my wish to make a robust note-detector, and aubio's onset detection may put me on a track.

Katja

so anyone can share with me the 32 bit version of this library? I would be grateful as I really need it on a macosx.

Hi dio_333, attached is moog1's modified version compiled for OSX 10.5, 32 bit. Maybe it will work on your system, if you have 32 bit Pd.

Katja

http://www.pdpatchrepo.info/hurleur/aubio.pd_darwin.zip

thank you! it seems that one should also have the 32 bit of the aubio library and 32 bit versions of all its dependencies(fft3w,libsamplerate) in order to get this to work.

Ai, life is so complicated with those different bitnesses. I've collected all the files, aubio lib files are in /usr/local/bin/ and the other dependencies in /sw/lib/, assuming a fink install (probably /usr/lib/ otherwise).

If this does not work, you may rather download a 64 bit Pd 0.43 from the nightly builds, if you have the 64 bit aubio stuff compiled?

Katja

yes...I tried that but I get symbol not found. I will probably use some bonk onset detection and use aubio onset detection on a linux computer. thank you!

Seems I did not attach those files mentioned in post #29. Another try.

Edit: no it doesn't work, probably the .zip is too large (4.4MB), it is discarded or so. Sorry.

Katja

I see...could you sent it to my email if it's not too much.
dio_333
[at]
yahoo.com

thanks for sending me that! I think it also needed the libsamplerate and I finally decided to install everything with macports with 32 bit.

What I did was to edit the macports.conf file (in /opt/local/etc/macports/) and activate the build_arch line, seting it to 'build_arch i386'. One should also use this command "sudo port upgrade --enforce-variants whateverportlib +universal" for whatever libs for which macports claims that there are no 32bit versions.

of course, afterwards I copied the libs from /opt/local/lib to /usr/local/lib. if anyone knows a more elegant solution, please tell me.

anyway...thanks everyone, katja and moog especially!

I would mark this as SOLVED

one more thing...I realized why it was not working for 64bits when I always got the error undefined symbol aubio_onset_do. It was because in the aubio source code I have aubio_onset instead of aubio_onset_do. It seems that Paul modified this and I have a older copy of aubio source code on my computer, even if I thought that I have a new one.

I had to rebuild the aubioonset because the threshold inlet didn't do anything and it should call the function aubio_onset_set_threshold(x->o,f);
There are many things missing from this pd-aubio library...

I posted this just in case someone might need it someday if trying to compile the aubio library.

I am posting the modified aubioonset for high frequency content onsets and with the threshold functioning (0.001-0.9 values). you need to compile it of course but I can provide the mac/linux externals for 32 bits if anyone needs them.

important edit!
I deleted the file because it had delay in detecting the onsets because I was calling the function to set the threshold every time each sample was processed. the threshold should be set somewhere at the beginning, anyway not while processing the audio because it can introduce latency

if anyone is still interested, I managed to have the aubio library compiled in the external on macosx. I wrote another external in which you can change the detection method, window, hop sizes and thresholds as described in paul brossier's phd thesis.

the usage is
aubioOnset~ hfc 512 128 -70 0.5

http://www.pdpatchrepo.info/hurleur/aubioOnset~.pd_darwin