Musical Key Estimation of Audio Signal Based on Hidden Markov Modeling of Chroma Vectors

Geoffroy Peeters
DAFx-2006 - Montreal
In this paper, we propose a system for the automatic estimation of the key of a music track using hidden Markov models. The front-end of the system performs transient/noise reduction, estimation of the tuning and then represents the track as a succession of chroma vectors over time. The characteristics of the Major and minor modes are learned by training two hidden Markov models on a labeled database. 24 hidden Markov models corresponding to the various keys are then derived from the two trained models. The estimation of the key of a music track is then obtained by computing the likelihood of its chroma sequence given each HMM. The system is evaluated positively using a database of European baroque, classical and romantic music. We compare the results with the ones obtained using a cognitive-based approach. We also compare the chroma-key profiles learned from the database to the cognitive-based ones.