On Tue, Jan 29, 2008 at 02:51:11PM -0500, Albert Santoni wrote: > > The difference is, however, that our BPM detection still sucks > > compares > > to Pioneer's, so I think a prerequisite of a useable autolooper is a > > better Tap function which intelligently adjusts the beat marks and > > informs the beat marking engine how to do its job better for this > > track > > (in a way it'll remember next time!) rather than just updating the > > BPM. > > I haven't seen a BPM detection algorithm that fairs significantly > better than ours. If you want to improve the BPM tap, I'd invite you > to hack it. The way these BPM tap-assisted detection algorithms work > is they simply use the BPM tap guess to narrow the range of BPMs to > look at. > > Rough algorithm: > 1) Get user to tap to the BPM > 2) Smooth over their taps, do some standard deviation calculation > 3) Tell the BPM detection to search within +- std. dev. > 4) you're done
Seems about right, except I'd characterise 3 as "tell the *beat* detection to search within +/- x of the tap, and move existing beat marks where they are currently in the wrong place". This is the part that doesn't happen at the moment, the tap only feeds into the BPM calculation, not the beat detection. > Does Pioneer's algorithm work > better than Mixxx's _without_ any extra bpm-tap assist?) Yes. The BPM displays on the CDJs don't have a tap button. They are fast and very rarely wrong. The DJM does have a tap button but it's incredibly rare to have to use it. They work with all styles of music including breaks which I find[*] the mixxx algorithm frequently still has problems with. [*] Or did, last time I tried it which was when Micah was technically still working on it but I don't think he made any changes to the actual algorithm after that. Pioneer have led this field for many years. Other manufacturers put BPM counters on their mixers and CD decks but they just aren't as good (the Denon DN-X1500S was pretty good, but still struggled with certain types of music and had a tendancy to "flap"). I suspect even if Pio's algorithm could be reverse engineered, their patents probably have a few years left to run... Ben ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Mixxx-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mixxx-devel
