I've been experimenting with different looping algorithms for use with sinusoidal modeling synthesis. Typically, when one develops this kind of a synthesizer, one has separate sound files for the attack portion and the steady state portion of a sound, and the file for the steady state portion, i.e. the part that will be looped, is played very carefully so that it has a constant amplitude, pitch, brightness, and so on, because this makes the loop seamless. I had been hoping to get around the constraint that the sound material had to be carefully performed by analyzing the sound file and automatically identifying good loop points. So, I compared all the frames in a test sound file with the earth mover's algorithm, and I chose those frames with the smallest earth movers distance as loop points. I honestly must say though that I'm a bit disappointed with the results, because the loop points are still audible. Here's a sample:

http://www.gregkellum.com/temp/a4_multipleloops.mp3