HTM Learning Algorithm Implementation for learning and regenerating musical Sequences
This Project is trying to create a implementation of the HMT learning Algorithm (see www.numenta.org), which is able to recognise and model common, reoccuring pattern or structures in a given set of musical data. This sould then be used to 'rewrite' compareable musical sequences by combining the learned patterns in a new way. This way one could have all his/her favorite music (with similiar attributes) learned and modeled by the HTM and have it regenerate music that recombines ones favorites.
Music. This is what We all love and enjoy: But some of the most popular composers are not alive anymore. In millions of musical data-files, provided from various online sources we still have their work, their data, their musical patterns. Once this can be done, it could be expanded to be trained on every users favorite playlist and then generate structurally comparable songs on demand.
This is the goal of this project: Using data analysis tools and frameworks, I would like to extract the very quintessence out of the music, beginning with monophonic piano sequences, of great composers - like Mozart - and with these learned patterns then create new music following Mozarts principles. To do this, I have in mind using the NuPIC platform (see: http://numenta.org/ ) for intelligent data analysis build upon neuro-scientific discoveries. This hierarchically built framework of multiple sequence detectors are well suited to accomplish this task and would illustrate a beautiful way to combine three inspiring fields of our time: Data, Music and the Brain.
By combining these fields in this way, the "NuMozart"-Project aims to increase popular awareness of the three fields, as music is likely to draw a lot of attention to data-science and computational neuroscience respectively. Therefore, NuMozart opens up a lot of possibilities: Attention due to the 'recreation' of Mozart or other long gone Composers. Insight by applying a Cortex-like sturcutred neural network model to the musical data, which might create a better and deeper understanding of the computational operations that occur in the Brain when composing music, and various other possible uses (like theory-testing of the NuPIC learning theory).
Pascal Weinberger