Software Hears and Writes Music, Like a Personal Debussy
As PopSci explains, the software first figures out the spectral pattern of a given solo piece, which is used to create a harmonic dictionary. The dictionary is then transformed into a pattern algorithm, which is used to match sounds with notes using MIDI. In essence, it's like the Shazam iPhone app for conservatory students. Whereas most other transcription software programs are tuned to pick up on certain notes, this new device creates its own musical lexicon, depending on what it hears. Regardless of instrument, musician or recording conditions, the software can adapt its harmonic dictionary accordingly, and evolve over time.
So far, the system can only transcribe music played by one instrument, although researchers are optimistic that they'll eventually be able to incorporate more complex musical structures into the framework. Co-author Julio José Carabias points out, however, that even in its present state, the new software could greatly help musicologists to more closely analyze pieces, or to separate various audio sources. Even if you've never picked up a violin or tried to play a kazoo, you've still got to admit that the thought of software learning and writing music "by ear" is pretty incredible. [From: PopSci]