wav2xml

Take an MP3 file, extract the instruments performing the music (as well as the singing - we're dreaming anyway) and convert this information to sheet music in MusicML format for example. The Melisma Music Analyzer seems to be performing part of this function.

The first application of such a technology would be to find music by content: humming a tune to the system, which would then search for a similar melody in its database. Or input a partial mp3 file to find its reference. This is becoming a hot topic, as witnessed by the Google results. Existing systems are starting to emerge, like this one.

It is also interesting to think of the problem as related to its cousin, querying image by content. IBM already has commercialized technology performing this function. The idea would be to apply the same concepts, but adapted to the sound space of audio music rather than image space. In both cases, it is necessary to extract structural features from the data, which in some sorts represents the projection of the originating device's output onto the image or the mp3. The problem is that the data does not necessarily contain //explicit// structural information.