Any programmers?

[deathy]

If you could find an externally usable tool for automatically detecting BPM, then what you suggest is theoretically possible, but Traktor wouldn't work for it very easily, as it would be work to get it to do it without human intervention. I don't know of any command line or otherwise externally automatable BPM detection tools off the top of my head.

mostapha is right, though... the hardest part is determining the beat. The basic idea isn't too bad, you take an FFT with a high enough resolution, check the frequencies for where the different drums normally lie, and mark up where those frequencies are high.

However, the problem is that you get a lot of frequency cross-over all over the place, bass fighting the kick, singer fighting the snare, lead fighting the hats, etc. So, after you have the frequency domains determined, then you need to start looking for patterns in the data that looks like a beat... such as "In the kick range, we have thumps here, here, here, and here. That one seems out of place, so is probably the bassline, but the others suggest... etc. etc."

Now, the intelligence your program requires to be able to do this part has one really big advantage - the assumption of a single tempo. You may not know the tempo, but because your code assumes a single tempo, it can look for these patterns and be relatively accurate about it.

However, when you are actively looking for tracks where this tempo changes, the problem domain becomes larger. Your heuristics start to get a whole lot uglier. Slicing it up into pieces would work with limited success, but the accuracy you would enjoy at that point would be low enough that I am confident you would get a lot of false negatives.

That's not to say it can't be done, but to do it properly would require someone with a lot of audio programming knowledge.

[ZxZDeViLZxZ]

MIK will detect Bon and write it to tag or file name could pre process tracks in MIK then have a program just check the track slices against what the id3 tag is or name tag

[ImNotDedYet]

MIK will detect Bon and write it to tag or file name could pre process tracks in MIK then have a program just check the track slices against what the id3 tag is or name tag

That's the crux of the issue though...is MIK's beat detection correct or is it an average of all the different slices? Were they able to accurately determine the BPM?

In order to get this functionality, you have to have something that can relatively accurately determine BPM, and so far software apps have proven they can get close, in the ballpark, maybe even fairly precise in some instances. And fail miserably in others. Traktor also has this capability. But in my experience, it's made plenty of errors in that regard. Had it not, no one would have to go in and manually enter beat grids.

[ZxZDeViLZxZ]

I don't enter beat grids lol... Yes but if a ballpark bpm is determined and definite then you cut the 30second slices grab a bpm detection of slices then determine and average then look at the average determine is it is in fact different from what the id3 tag shows if within 3% then it would assume it's the same if its over 3% then it would then analyZe each slice and compare against average and the id3 tag if it's still out by 3% it can be determined that it is a wandering bpm track

[ZxZDeViLZxZ]

This does a few important things it grabs from two bpm algorithms and then compares it then compares 30second slices which can be anywhere from 8 to 32 slices which increases your chances of it making a correct descion... You could also implement an open source metronome library that will then take the determined bpm of a track and plug into the metronome the bpm and then find say a frequency between 50hz and 200hz and analyze the spikes against the metronome hits

[ZxZDeViLZxZ]

Honestly I think software bpm detection is poor because most the time a close guess is good enough... Thinking back to vinyl you really likely never were 100% but 85% was good enough for that minute transition and the lower your accuracy the quicker your mixes have to be

[Patch]

The way I think about it is - if software is written to only find the tracks that have a large tempo change in 'em, I'd be happy to manually set the initial bpm anyway. Out of anyones library, there aren't gonna be a huge number of transition type tracks (tracks with a significant tempo change).

These tracks aren't usable for the "sync'ers" anyway - but if you can beatmatch, and know the tempo AFTER the change, you can at least select a track that is in the right ball park tempo.

[SlayForMoney]

i got the info that there was a program out there that put out a graph of how bpm changed over time but it was on win98. supposedly AUTO BPM but google search didn't give any results

[ZxZDeViLZxZ]

http://wiki.hydrogenaud.io/index.php...yser_(foo_bpm) Maybe a help

[mostapha]

I've been doing a bit of research and found some libraries to do the DSP. I might take a crack at it if I have enough downtime over the next few days.

But, I still think it'd be easier to just listen to the songs.