Secrets of Tempo-Detection Software

A while back, I wrote about some slick tempo-detection software. A week later, Erica Sadun reported on a utility she was using, and complained about how long it took to analyze songs.

The developer of that program just wrote to explain that compressed music files like MP3s must be expanded before they can be analyzed, and that takes time. Curious, I contacted Aaron Higgins of MixMeister Technologies, which makes some powerful tempo-detection programs for Windows. He confirmed that surprising statement, and explained why:

Yes, it's true that you need to decompress before analyzing. There are several tricks you can use to minimize the workload, like only decoding a few short clips of the track as opposed to the whole thing. Alternatively, it might be possible to partially decompress the data since you don't need to reconstruct the entire frequency band, just the most prominent stuff. But, this optimization requires that you actually write your own decompression routines, which I don't believe is worth it. In practice, it is just easier to generically decode everything and then write your analyzing routines assuming decompressed audio. That is what we do. If you want to measure the amount of time spent decompressing, you could try analyzing the same song twice, once as a wave and again as an MP3.

Given how useful knowing the tempo of a song can be, it would be cool if ID3 tags could store a highly accurate tempo map (not just an overall tempo). That would enable some exciting synchronization techniques, such as changing slides or movie clips on a beat or quickly organizing songs into flowing playlists.

Speaking of which, Higgins also mentioned that MixMeister just released a DJ sequencer called Fusion, which looks to take creative synchronization to a new level.