Optical Music Recognition is a field of research that investigates how to computationally decode music notation from images. Despite the efforts made so far, there are hardly any complete solutions to the problem. In this work, we study the use of neural networks that work in an end-to-end manner. This is achieved by using a neural model that combines the capabilities of convolutional neural networks, which work on the input image, and recurrent neural networks, which deal with the sequential nature of the problem. Thanks to the use of the the so-called Connectionist Temporal Classification loss function, these models can be directly trained from input images accompanied by their corresponding transcripts into music symbol sequences. We also present the Printed Images of Music Staves (PrIMuS) dataset, containing more than 80,000 monodic single-staff real scores in common western notation, that is used to train and evaluate the neural approach. In our experiments, it is demonstrated that this formulation can be carried out successfully. Additionally, we study several considerations about the codification of the output musical sequences, the convergence and scalability of the neural models, as well as the ability of this approach to locate symbols in the input score.

Music genre meta-data is of paramount importance for the organization of music repositories. People use genre in a natural way when entering a music store or looking into music collections. Automatic genre classification has become a popular topic in music information retrieval research both with digital audio and symbolic data. This work focuses on the symbolic approach, bringing to music cognition some technologies, like the stochastic language models, already successfully applied to text categorization. The representation chosen here is to model chord progressions as n-grams and strings and then apply perplexity and Naive Bayes classifiers in order to model how often those structures are found in the target genres. Some genres and sub-genres among popular, jazz, and academic music have been considered and the results at different levels of the genre hierarchy for the techniques employed are presented and discussed.

Abstract. In a music recognition task, the classification of a new melody is often achieved by looking for the closest piece in a set of already known prototypes. The definition of a relevant similarity measure becomes then a crucial point. So far, the edit distance approach with a-priori fixed operation costs has been one of the most used to accomplish the task. In this paper, the application of a probabilistic learning model to both string and tree edit distances is proposed and is compared to a genetic algorithm cost fitting approach. The results show that both learning models outperform fixed-costs systems, and that the probabilistic approach is able to describe consistently the underlying melodic similarity model.

No abstract

No abstract

Abstract. Music genre meta-data is of paramount importance for the organization of music repositories. People use genre in a natural way when entering a music store or looking into music collections. Automatic genre classification has become a popular topic in music information retrieval research. This work brings to symbolic music recognition some technologies, like the stochastic language models, already successfully applied to text categorization. In this work we model chord progressions and melodies as n-grams and strings and then apply perplexity and naïve Bayes classifiers, respectively, in order to assess how often those structures are found in the target genres. Also a combination of the different techniques as an ensemble of classifiers is proposed. Some genres and sub-genres among popular, jazz, and academic music have been considered. The results show that the ensemble is a good trade-off approach able to perform well without the risk of choosing the wrong classifier.

Genetic-based composition algorithms are able to explore an immense space of possibilities, but the main difficulty has always been the implementation of the selection process. In this work, sets of melodies are utilized for training a machine learning approach to compute fitness, based on different metrics. The fitness of a candidate is provided by combining the metrics, but their values can range through different orders of magnitude and evolve in different ways, which makes it hard to combine these criteria. In order to solve this problem, a multi-objective fitness approach is proposed, in which the best individuals are those in the Pareto-optimal frontier of the multi-dimensional fitness space. Melodic trees are also proposed as a data structure for chromosomic representation of melodies and genetic operators are adapted to them. Some experiments have been carried out using a graphical interface prototype that allows one to explore the creative capabilities of the proposed system. An Online Supplement is provided where the reader can find some technical details, information about the data used, generated melodies, and additional information about the developed prototype and its performance.

Genetic-based composition algorithms are able to explore an immense space of possibilities, but the main difficulty has always been the implementation of the selection process. In this work, sets of melodies are utilized for training a machine learning approach to compute fitness, based on different metrics. The fitness of a candidate is provided by combining the metrics, but their values can range through different orders of magnitude and evolve in different ways, which makes it hard to combine these criteria. In order to solve this problem, a multi-objective fitness approach is proposed, in which the best individuals are those in the Pareto-optimal frontier of the multi-dimensional fitness space. Melodic trees are also proposed as a data structure for chromosomic representation of melodies and genetic operators are adapted to them. Some experiments have been carried out using a graphical interface prototype that allows one to explore the creative capabilities of the proposed system. An Online Supplement is provided where the reader can find some technical details, information about the data used, generated melodies, and additional information about the developed prototype and its performance.