Transients and latent spaces
'Transients and Latent Spaces' is my first work exploring the applications of deep learning and generative AI to concert music composition. Two deep models from Google's Magenta project were used as part of the compositional process: drums_rnn and MusicVAE.
I trained the drums_rnn model on drum patterns based on rhythms from The Sacrificial Dance, the final movement of Stravinsky's Rite of Spring which brings to the forefront many of the rhythmic devices employed by Stravinsky throughout the rest of his piece. These drums patterns were obtained by manually adding an accompanying drum part to sections of this movement, trying to match the rhythmic accenting of the original work as accurately as possible. This model was used to generate various sections of material which functioned as structural 'pillars' of the piece.
The MusicVAE model, trained on a massive corpus of drum beats from popular music, was used to interpolate between these 'pillars' of Stravinsky-influenced material generated by drums_rnn, in a way that was smooth-sounding yet also used natural and idiomatic beat patterns.
Studio recording of my piece for Beatboxer, 'Transients and Latent Spaces'.
Beatbox - Omar Peracha