This paper investigates the use of musical priors for sparse expansion of audio signals of music on overcomplete dictionaries taken from the union of two orthonormal bases. More specifically, chord information is used to build structured model that take into account dependencies between coefficients of the decomposition. Evaluation on various music signals shows that our approach provides results whose quality measured by the signal-to-noise ratio corresponds to state-of-the-art approaches, and shows that our model is relevant to represent audio signals of Western tonal music and opens new perspectives.