It’s pretty scary reading about how talented Hilary Hahn is, how much she achieved at such a young age. But, as this interview with Guernica shows, it’s all about the music, and Bach in particular.
Hilary Hahn: Entering the sublime
When I play the solo repertoire, the way Bach writes is pretty progressive. I believe he would be considered somewhat experimental even by today’s standards. He was a tonal composer as opposed to atonal or twelve-tone and he used acoustic instruments because he had nothing else; but when I listen to the progressive aspect of his music, I feel like he’s a master of disguise. You are going with him in one direction—you get there, but you realize you are not where you thought you were. You look around and notice a door, but when you get to the door, it’s a wall. You look around and notice that the floor you’re standing on is a trap door. You go down the trap door to what you think is the basement, but it’s the attic. Bach feels a lot like that—a really interesting fun house. The dimensions are different from a distance than when you get up close.
Hilary Hahn – J.S. Bach: Sonata for Violin Solo No. 1 in G Minor, BWV 1001 – 4. Presto
Otherworldly talent. But as this video shows, she is human.
Hilary Hahn does the Ling Ling Workout
I love this Google Doodle, though even Bach can’t rescue my appalling lack of musical ability!
Google’s first AI-powered Doodle is a piano duet with Bach
Starting on March 21st, you’ll be able to play with the interactive Doodle, which will prompt you to compose a two-measure melody or pick one of the pre-existing choices. When you press the “Harmonize” button, it will use machine learning to give you a version of your melody that sounds like it was composed by Bach himself.
Various Google teams were involved in this project, including Google Magenta. There is an incredible amount of detail about the technologies behind the Bach harmonies on their own site.
Coconet: the ML model behind today’s Bach Doodle
Coconet is trained to restore Bach’s music from fragments: we take a piece from Bach, randomly erase some notes, and ask the model to guess the missing notes from context. The result is a versatile model of counterpoint that accepts arbitrarily incomplete scores as input and works out complete scores. This setup covers a wide range of musical tasks, such as harmonizing melodies, creating smooth transitions, rewriting and elaborating existing music, and composing from scratch.
I cannot begin to understand what’s going on there, but it sounds good.