Beat Detection Algorithms, Part 2

mziccard.me

20 points·snickmy··6 comments

He's extracting BPM info, but not phase. You need phase info to sync two tracks for mixing.

There are some "automatic DJ" programs around, such as Virtual DJ and Softonic, but automatic mixing hasn't yet become a common feature in music players. It should be.

There was a time back in the late 90's when I wanted to write my own music/DJ apps, but I had no idea how to do things like graph audio or do bpm detection for example... or even where to look for such info.

Nowadays, the info comes to me. Too bad I can't send these links back to my younger self.

Wonder how this algo compares to Scheirer's approach. ( http://www.music.mcgill.ca/~hockman/coursework/MUMT_614/sche... )

And what if there is not a constant tempo?

If you know the score, you can more easily find beats and tempo (and do score following). For instance: http://repmus.ircam.fr/antescofo

You could probably split the song into frames, detect the BPM of each frame, and plot the changes in BPM over time. This sort of thing would be pretty for pre-rendering metadata about the tempo, which could be useful in DJ software, or big music databases, stuff like that.

The IRCAM research you link to looks really interesting, and I think it would be useful for other purposes, like real-time performance with automatic accompaniment.

Yes, but you have always a duality problem between tempo and beat, i.e. are they two eight notes at that tempo, or two quarter notes at twice quicker tempo...

You can assume that the most prevalent duration is a quarter note, but it is not always what would be on the score.

But it's more a problem related to transcription than to DJing (where the representation of the beat as a whole note, a quarter note, or anything else is eventually not important, I think).

Reply on news.ycombinator.com