Google does some weird things in its research that occasionally transform into cool items that we can tinker with on the web. Presently its Magenta research arm has made an AI that can transform anybody into a piano virtuoso. The creation is called Piano Genie and it’s an AI program that enables anybody to improvise on the piano by beating on eight catches associated with a genuine player piano. Piano Genie was motivated by Guitar Hero, an amusement that was madly prominent a couple of years back and rearranged playing the guitar, transforming playing music into a game.
At the point when group Magenta made Piano Genie, it took an unexpected course in comparison to Guitar Hero, a game acclaimed for enabling players to play pop and shake tunes composed by surely understood performers. Piano Genie is intended to enable clients to make another song on the fly utilizing an AI program and a goliath dataset of established piano music. The AI can foresee takes note of that ought to pursue each other in a way like how your cell phone can anticipate what word may pursue next while making an instant message.
One of the makers of Piano Genie is Chris Donahue, an assistant at Google Magenta, and one of the group of three taking a shot at the venture. The other two individuals from the group incorporate Ian Simon and Sander Dieleman; Dieleman is a veteran of Google’s DeepMind. Donahue says that Piano Genie can ad lib note to note to give the client a sentiment of power over the music. He said when you are playing Piano Genie “at times, it will feel like it’s kind of perusing your brain and play the correct note you are proposing to.” He likewise takes note of that occasionally the AI will “resist you” however that despite everything it accomplishes something sensible. This is the reason the framework is called Genie says Donahue, since you can wish for what you need, however, what you get isn’t generally what you requested.
The machine learning parts of the venture are developed from normal AI components; the principal segment is a repetitive neural system that is great at figuring out how to impersonate successive information like composition and music. That neural system was prepared with a dataset of piano music taken from a global rivalry called the piano-e-rivalry. This was an especially decent dataset for the task in light of the fact that the opposition records the notes played, as well as speed too, converting into timbre and volume. That information was utilized as the fundamental preparing information for Piano Genie and manufactured the prescient model of what piano notes pursue one another.
The Piano Genie group at that point outlined a couple of encoders that could consolidate that yield into a configuration suited to their eight-catch controller by compacting the notes from each of the 88 keys on a standard piano to work with eight catches. The last bit was joining what they had made to oneself playing piano found in the recordings.
Source:
- https://hothardware.com/news/piano-genie-ai
- https://magenta.tensorflow.org/pianogenie