A lossy audio encoding format written by Andree Buschmann originally based on the theories behing MPEG-1 Layer 2 (The predecessor to MP3). It is prided to be the highest quality lossy scheme out there, while still remaining fairly fast in its encoding speed.

The algorithm works on the same principle as MP2. That is, a signal mask is generated for each subband (usually a few khz wide) that is based on the Absolute Threshold of Hearing (ATH) . The resulting signal that is audible is then differenced from this base ATH and passed on to the huffman encoder.

Its superiority to other lossy algorithms comes from its simplicity. It does not use fast fourier transforms to encode its signal. Instead it takes advantage of temporal masking and other tricks to reduce the amount of raw data that needs to be encoded (all manifested in the ATH).

The license that musepack has eventually come to is unusual. Originally, the author was just writing it for a class thesis. Then when some random people off of DALnet IRC network got ahold of it, the interest grew larger. Smartly the author never released any source code during this time. Shortly afterwards, he was approached with commercial interest by a large german corporation that I cannot disclose. However, he has released the source code for the decoder, and binaries for nearly every platform imaginable exist. Thus, the license it has arrived at is a "use this for whatever you want personally, but commercial use without talking to me is a no-no" one.

The original site is still available at http://www.personal.uni-jena.de/~pfk/MPP/audiocoder_english.html