- Log in to post comments
Question
How can we easily identify piano pieces by era? It is tedious to go piece by piece composer by composer and still get conflicting information from places like google. If you want to section your music on your phone or sheet music, there is really no elegant, easy solution, especially for lesser known pieces or pieces with contested composers.
Solution
Using machine learning, we can build a model that can easily recognize piano pieces by musical eras (Baroque, classical, romantic, etc.) based on structures and patterns in the music. This would massively help efficient sorting for large collections of music.
Method
- With a combination of MIDI datasets like MAESTRO, Kunstderfuge, and others like them, we can compile a dataset of piano compositions spanning the multiple major musical eras.
- We can use factors like: Tempo, note density, key changes, rhythmic complexity, and harmonic intervals using MIDI analysis tools.
- Then label each piece by musical style based on the composer and historical context.
- Train models on this and compare and evaluate performance with pieces that were not used.
Progress
So far, we have collected and found some MIDI databases that can be used for testing.