22 Mar
2004
22 Mar
'04
9:04 a.m.
Hello,
I'm working on a simple command line driven bpm
counter for mp3 files. It's very straight-forward.
I'm trying to use a 4-beat phase shifting technique,
but for some reason my code is returning the same
exact value for each file. Obviously I have a stupid
math error, or haven't thought things through enough.
If someone wants to look at this and tell me what they
think I did wrong, I'd appreciate it :) The code is
pretty clean (just wrong) and well commented. You'll
need to have mpg123 installed for it to (not) work.
Thanks very much!
=====
Jay Dolan
Software Engineer, Systems Analyst
Windmill Cycles, Inc.
508.999.4000
__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html
22 Mar
22 Mar
11:46 a.m.
Jay Dolan wrote:
Hello,
I'm working on a simple command line driven bpm
counter for mp3 files. It's very straight-forward.
I'm trying to use a 4-beat phase shifting technique,
but for some reason my code is returning the same
exact value for each file. Obviously I have a stupid
math error, or haven't thought things through enough.
If someone wants to look at this and tell me what they
think I did wrong, I'd appreciate it :) The code is
pretty clean (just wrong) and well commented. You'll
need to have mpg123 installed for it to (not) work.
Thanks very much!
Try this?
(I haven't).
Simon Jenkins
(Bristol, UK)
=====
Jay Dolan
Software Engineer, Systems Analyst
Windmill Cycles, Inc.
508.999.4000
__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html
1:46 p.m.
--- Simon Jenkins <sjenkins(a)blueyonder.co.uk> wrote:
Try this?
(I haven't).
Simon Jenkins
(Bristol, UK)
if(k == g_minBpm || diff < lowestDiffFound){
bpm = k;
lowestDiffFound = diff;
}
Definitely closer! I'm embarassed to admit how long
I've looked at this code and didn't see that I was
initializing lowestDiffFound (was previousDiff) to 0,
doh!
It's now returning 120.0 (g_minBpm) for every file,
tho. So there's still a problem somewhere. Perhaps
in phaseShift? I've re-attached the improved version.
Thank you for your help.
=====
Jay Dolan
Software Engineer, Systems Analyst
Windmill Cycles, Inc.
508.999.4000
__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html
9:14 p.m.
Jay Dolan wrote:
--- Simon Jenkins <sjenkins(a)blueyonder.co.uk>
wrote:
ok:
~ Downsampling by 50??? That's *got* to require some
filtering. In keeping with your need for speed - not to mention
my lack of DSP guru-head - i've implemented the crudest
imagineable.
~ fseek()-ing the pipe wasn't working. I've removed the
first fseek(), but that means that g_startTime is no longer
honoured and must be left at zero. The (ahem) "filter" i
added means the second fseek() is no longer required.
~ A couple of math corrections.
~ Widened expected bpm range: It now looks for 100 - 199 bpm.
You should probably narrow that back down again if you know
your source is in a narrower range.
I actually ran it a couple of times and it seems to produce plausible
results now though i don't have the exact bpms for the mp3s i tried.
Also, my test mp3s contained looped samples and/or programmed
drum parts. I'm not sure this algorithm will work so well (or at
all for that matter) on less regular material.
Simon Jenkins
(Bristol, UK)
Try this?
[...]
Definitely closer! I'm embarassed to admit how long
I've looked at this code and didn't see that I was
initializing lowestDiffFound (was previousDiff) to 0,
doh!
It's now returning 120.0 (g_minBpm) for every file,
tho. So there's still a problem somewhere. Perhaps
in phaseShift? I've re-attached the improved version.
Thank you for your help.
11:55 p.m.
--- Simon Jenkins <sjenkins(a)blueyonder.co.uk> wrote:
ok:
~ Downsampling by 50??? That's *got* to require
some
filtering. In keeping with your need for speed - not
to mention
my lack of DSP guru-head - i've implemented the
crudest
imagineable.
~ fseek()-ing the pipe wasn't working. I've removed
the
first fseek(), but that means that g_startTime is no
longer
honoured and must be left at zero. The (ahem)
"filter" i
added means the second fseek() is no longer
required.
~ A couple of math corrections.
~ Widened expected bpm range: It now looks for 100 -
199 bpm.
You should probably narrow that back down again if
you know
your source is in a narrower range.
I actually ran it a couple of times and it seems to
produce plausible
results now though i don't have the exact bpms for
the mp3s i tried.
Also, my test mp3s contained looped samples and/or
programmed
drum parts. I'm not sure this algorithm will work so
well (or at
all for that matter) on less regular material.
Simon Jenkins
(Bristol, UK)
Simon,
Thanks so much for lending a hand :) I was working on
it too tonight, and received your email about 30
seconds after I got it working, hehe. I found that
while it's not "supported", the fseek works fine for
me. The man pages for fseek say that it can't be used
on pipes opened for /writing/. Obviously we're just
reading from a pipe..and while errno is still set to
EPIPE, it works here. However, I don't like hacks,
and so I'll replace all piping / reliance on mpg123
with calls to libmad.
You are dead on with the downsampling issue. With a
downsampleFactor of 1 or 2 the program works great for
me now - 50 was way too high. I know jack sh!t about
filtering..or PCM in general. But the program, while
slower, is now at least as accurate as my Numark
CDN-88's Beatkeeper III.
Again, I'm really greatful for your help. I'll attach
a completed version soon, hopefully. Take care,
=====
Jay Dolan
Software Engineer, Systems Analyst
Windmill Cycles, Inc.
508.999.4000
__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html
23 Mar
23 Mar
4:51 a.m.
On Mon, Mar 22, 2004 at 07:49:49 -0800, Jay Dolan wrote:
You are dead on with the downsampling issue. With a
downsampleFactor of 1 or 2 the program works great for
me now - 50 was way too high. I know jack sh!t about
filtering..or PCM in general. But the program, while
slower, is now at least as accurate as my Numark
CDN-88's Beatkeeper III.
You could use SRC to do the downsampling http://www.mega-nerd.com/SRC/
I think therers also a wavelets function that can do correlation of audio
more effeciently than a linear search, but I might have just dreampt it :)
- Steve
7:02 a.m.
On Tue, Mar 23, 2004 at 08:45:15AM +0000, Steve Harris wrote:
I think therers also a wavelets function that can do
correlation of audio
more effeciently than a linear search, but I might have just dreampt it :)
Filter, take amplitude, downsample, FFT, power spectrum, inverse FFT.
This gives the autocorrelation. Select the highest peak.
--
FA
24 Mar
24 Mar
1 p.m.
--- Steve Harris <S.W.Harris(a)ecs.soton.ac.uk> wrote:
You could use SRC to do the downsampling
http://www.mega-nerd.com/SRC/
I think therers also a wavelets function that can do
correlation of audio
more effeciently than a linear search, but I might
have just dreampt it :)
- Steve
Thanks, Steve. I ended up just averaging every n
samples to do the downsampling. This way allowed me
to use a default downsampling factor of 1/10, which
isn't bad.
The following file types are now supported: Mp3, Ogg,
Wav, and Raw. The program can also be piped to over
stdin, so long as the piped input is raw.
Requires sox and mpg123. You can grab it at
http://jaydolan.com by clicking on the "etc" page.
It's GPL'ed. Feedback/criticism is welcome. Thanks
for your help, guys :)
=====
Jay Dolan
Software Engineer, Systems Analyst
Windmill Cycles, Inc.
508.999.4000
__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html
7608
days inactive
7610
days old
linux-audio-dev@lists.linuxaudio.org
7 comments
4 participants
participants (4)
-
Alfons Adriaensen -
Jay Dolan -
Simon Jenkins -
Steve Harris