Re: [Audacity-devel] FW Re: Tempo detector

[Gale A. <ga...@au...>]

Isaac, 

Thanks, I'm forwarding to -devel again, with reply to set to -devel ,
so that others can read and reply if they wish. 

I did try your app. on an arbitrary 30 seconds pop sample with a 
pretty pronounced beat. The beat.ny label output was very 
inaccurate at default setting, missing many beats but bunching  
three or more adjacent labels at several beats.   

There was a parse error "30.106000 B" which is the location 
and name of the last label.

Your output says "Finded  22 Beats" which is the number of labels.

"0 beats differentials removed (0% from total)"

"Tempo detected 268.5 match percent 57.3749"

The actual BPM is 84, measured as near as makes no difference here:
http://www.all8.com/tools/bpm.htm

Steve's bpm-count.ny, based on a pretty good guess at the threshold, 
comes up with 85 BPM if you guess the threshold at -5 dB, but 162 BPM 
if you guess it at -6 dB. 

"Aubio Beat Tracker" ( http://vamp-plugins.org/download.html ) gets 
the beats really spot on at default settings, as it happens. However in
Audacity it doesn't actually display the BPM calculation which I 
assume it does in a full VAMP host that offers more GUI than our 
label track.

 


Gale 


| From Isaac Pascual <isa...@ya...> 
| Fri, 9 Sep 2011 17:40:22 +0100 (BST)
| Subject: FW Re: Tempo detector
> I thing to release my work with GPL2 license. There are my first real
> contribution to comunity...
> 
> Im a C/C++ programer and Nyquist is little confusing for me. I read 
> de bpm-count.ny that you link but I thing that this is very easy and 
> probably output bad results... My solution is more acurate because 
> isnt global it is local.
> 
> Maybe one nyquist programmer can port my C code to nyquist
> pluguin...
> I link to you a packet with the code an FEW examples.
> Probably its necessay a exaustive test
> 
> To compile: make
> To execute: ./main label.txt
> label.txt can obtained from Audacity:
>   - Make a track with the beat finder result
>   - Export this track like txt (File menu)
> 
> Bye
> 
> --- El vie, 2/9/11, Gale Andrews <ga...@au...> escribió:
> 
> > De: Gale Andrews <ga...@au...>
> > Asunto: FW Re: Tempo detector
> > Para: "Isaac Pascual" <isa...@ya...>
> > CC: fee...@au..., aud...@li...
> > Fecha: viernes, 2 de septiembre, 2011 19:11
> > 
> > Isaac, 
> > 
> > Thanks for writing. We are always glad to hear from new
> > programmers.
> > 
> > I'm Ccing this to our developers' list (audacity-devel). It
> > would be an 
> > idea to subscribe to that list which you can do here:
> > https://lists.sourceforge.net/lists/listinfo/audacity-devel
> > 
> > We are fairly strictly in feature freeze at the moment
> > until such time 
> > as we can release a new 2.0 stable version of Audacity.
> > Small 
> > exceptions are possible if a feature is small, not invasive
> > and has
> > no stability implications. An improved Beat Finder could
> > possibly 
> > qualify, though the current Beat Finder is a Nyquist
> > plug-in, so 
> > can be more easily isolated from Audacity:
> > http://audacity.sourceforge.net/help/nyquist
> > 
> > Have you looked at Steve Daulton's attempt to improve the
> > detection 
> > in the Nyquist Beat Finder:  
> > http://forum.audacityteam.org/viewtopic.php?f=42&t=59071
> > ?
> > 
> > One limitation of that current improvement is that labels
> > are not 
> > produced. What form does the output in your program take? 
> > 
> > Also how is your work licensed? Is it available to test
> > now? 
> > 
> > 
> > Thanks, 
> > 
> > 
> > 
> > 
> > Gale 
> > 
> > 
> > 
> > | From Isaac Pascual <isa...@ya...>
> > 
> > | Fri, 2 Sep 2011 17:00:34 +0100 (BST)
> > | Subject: Tempo detector
> > > Im a C++ programer and Matematic. Im programming a
> > tempo detector 
> > > based on your actual Beat Finder. It is working now
> > but is a external
> > > program. You need to export the beat track from
> > Audacity and put in the 
> > > input of the tempo detector. I think that is very easy
> > to put it on the 
> > > Audacity's code because is very little and fast.
> > > 
> > > The tempo detector is based on a statistical study of
> > beat diferences. The
> > > return of the program is like:
> > > 30(15%), 60(34%), 90(5%)
> > > In this case probably the correct tempo is 60 but is
> > possible that tempo are 
> > > a multiple or a divisor
> > > 
> > > I think that this tool is very useful for musicians. I
> > do it because I need it.
> > > There are some possiblities to put this code on future
> > releases?
> > > 
> > > Maybe are interesting to make a test with a lot of
> > songs and see the 
> > > correctness of the algorithm.
> > 
> >