Most automatic chord recognition systems follow a standard approach combining chroma feature extraction, filtering and pattern matching. However, despite much research, there is little understanding about the interaction between these different components, and the optimal parameterization of their variables. In this paper we perform a systematic evaluation including the most common variations in the literature. The goal is to gain insight into the potential and limitations of the standard approach, thus contributing to the identification of areas for future development in automatic chord recognition. In our study we find that filtering has a significant impact on performance, with self-transition penalties being the most important parameter; and that the benefits of using complex models are mostly, but not entirely, offset by an appropriate choice of filtering strategies. 1.

An interesting problem in music information retrieval is how to combine the information from different sources in order to improve retrieval effectiveness. This paper introduces an approach to represent a collection of tagged songs through an hidden Markov model with the purpose to develop a system that merges in the same framework both acoustic similarity and semantic descriptions. The former provides content-based information on song similarity, the latter provides context-aware information about individual songs. Experimental results show how the proposed model leads to better performances than approaches that rank songs using both a single information source and a their linear combination. 1.

This paper presents probabilistic n-gram models for symbolic chord sequences. To overcome the fundamental limitations in conventional models—that the model optimality is not guaranteed, that the value of n is fixed uniquely, and that a vocabulary of chord types (e.g., major, minor, ···)is defined in an arbitrary way—we propose a vocabulary-free infinity-gram model based on Bayesian nonparametrics. It accepts any combinations of notes as chord types and allows each chord appearing in a sequence to have an unbounded and variable-length context. All possibilities of n are taken into account when calculating the predictive probability of a next chord given a particular context, and when an unseen chord type emerges we can avoid out-of-vocabulary error by adaptively evaluating the 0-gram probability, i.e., the combinatorial probability of note components. Our experiments using Beatles songs showed that the predictive performance of the proposed model is better than that of the state-of-theart models and that we could find stochastically-coherent chord patterns by sorting variable-length n-grams in a line according to their generative probabilities. 1.