We used Independent Component Analysis (ICA) with sparse coding to analyze music spectral sequences. We modelled an audio spectrum as an approximate mixture of the spectra of individual notes, using our ICA approach to `unmix` this to find the individual notes and note spectra. Notes are assumed to be approximately independent, and sparse (mostly off). Results on synthesized harpsichord music are encouraging, producing an approximate piano-roll transcription, and a passable rendition of the original music when resynthesized. We are currently working to extend and improve this through the use of temporal information of note activities and to handle more complex timbral behaviour.
(See document for exact affiliation information.)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
