Musical instrument recognition using ICA-based transform of features and discriminatively trained HMMs

In this paper, we describe a system for the recognition of musical instruments from isolated notes or drum samples. We first describe a baseline system that uses mel-frequency cepstral coefficients and their first derivatives as features, and continuous-density hidden Markov models (HMMs). Two improvements are proposed to increase the performance of this baseline system. First, transforming the features to a base with maximal statistical independence using independent component analysis can give an improvement of 9 percentage points in recognition accuracy. Secondly, discriminative training is shown to further improve the recognition accuracy of the system. The evaluation material consists of 5895 isolated notes of Western orchestral instruments, and 1798 drum hits.