Automatic Chord Estimation

Automatic Chord Estimation is the task of extracting or transcribing a sequence of chords from an audio music recording. The reported accuracy of the task has improved significantly in recent years, with deep learning models achieving accuracy rates of approximately 80-85%. How- ever, these approaches are computationally expensive, re- quire large datasets, and often act as a black box.

I proposed a computationally efficient and streamlined probabilistic approach to ACE using data- driven Hidden Markov Models that incorporate musical contexts. My model learns transition probabilities directly from annotated data, conditioned specifically on musical key and meter. This allows the system to model realis- tic harmonic behaviors, such as the likelihood of cadences within a key and the tendency of chord changes to occur on strong beats.