Additional layers would find patterns on the scale of whole choruses and verses.
#Master sun song windows#
The next layer might look at windows the length of a measure and be able to identify repeated riffs and the like. This would be the layer that judges rhythm, by taking notice of any deviance in their positions. I imagine a deep neural net focused on spectral analysis that on the lowest level is looking at a period of time equivalent to that of maybe a beat, determining where in that window an onset is likely to occur. specific moments in time, a more comprehensive machine learning approach could be implemented. Instead of relying entirely on analyzing rhythm, i.e. This could be accomplished by reworking the algorithm that judges what parts of the music are best. Although more difficult to get working correctly, this would be more streamlined and straightforward from a user's perspective, as well as address certain idiosyncrasies such as what happens when MIDI notes have long attacks or when dealing with glissando in vocals or other instruments.
#Master sun song upgrade#
One major upgrade would be to decouple The Amanuensis from its reliance on MIDI. Playtesters wanted! Version 2.0Īt this point perhaps one use of an open-source effort might be in creating a brand-new version 2.0 from the ground up.
0 kommentar(er)
