Chord recognition using Gaussian mixture model

摘要

In this paper, the method of recognizing the chord of a musical piece using a Gaussian mixture model (GMM) is proposed, and an experiment examination result is reported. The input sound signal from frame division is divided for every octave using filterbank, and the power spectrum corresponding to 12 sound name is extracted as a 12-dimensional vector from each octave. It asks for mixed dignity, an average vector, and a covariance matrix from this vector; and GMM is formed from these. Additionally, by using N-gram model together, the model is in consideration of chord progression. And above, it can be adapted to non-learned range, and the method of the case of a complicated composition sound on real time into consideration in the future is proposed.