This paper is concerned with algorithms for prediction of discrete sequences over a finite alphabet, using variable order Markov models. The class of such algorithms is large and in principle includes any lossless compression algorithm. We focus on six prominent prediction algorithms, including Context Tree Weighting (CTW), Prediction by Partial Match (PPM) and Probabilistic Suffix Trees (PSTs). We discuss the properties of these algorithms and compare their performance using real life sequences from three domains: proteins, English text and music pieces. The comparison is made with respect to prediction quality as measured by the average log-loss. We also compare classification algorithms based on these predictors with respect to a number of large protein classification tasks. Our results indicate that a “decomposed” CTW (a variant of the CTW algorithm) and PPM outperform all other algorithms in sequence prediction tasks. Somewhat surprisingly, a different algorithm, which is a modification of the Lempel-Ziv compression algorithm, significantly outperforms all algorithms on the protein classification problems.

All musical pieces are similar, but some are more similar than others. Apart from serving as an infinite source of discussion (‘‘Haydn is just like Mozart—No, he’s not!’’), such similarities are also crucial for the design of efficient music information retrieval systems. The amount of digitized music available on the Internet has grown dramatically in recent years, both in the public domain and on commercial sites; Napster and its clones are prime examples. Web sites offering musical content in some form like MP3, MIDI, or other, need a way to organize their wealth of material; they need to somehow classify their files according to musical genres and subgenres, putting similar pieces together. The purpose of such organization is to enable users to navigate to pieces of music they already know and like, but also to give them advice and recommendations (‘‘If you like this, you might also like...’’). Currently, such organization is mostly done manually by humans, or based on patterns in the purchasing behaviors of customers. However, some recent research has been examining the possibilities of automating music classification. A human expert, comparing different pieces of music with the goal of clustering similar works together, will generally look for certain specific similarities. Previous attempts to automate this process do the same. Generally speaking, they take a file containing a piece of music and extract from it various specific numerical features, related to pitch, rhythm, harmony, etc. One can extract such features using, for instance, Fourier transforms (Tzanetakis and Cook 2002) or wavelet transforms

This paper presents a representation for melodic segment classes and applies it to music data mining. Melody is modeled as a sequence of segments, each segment being a sequence of notes. These segments are assigned to classes through a knowledge representation scheme which allows the flexible construction of abstract views of the music surface. The representation is applied to sequential pattern discovery and to the statistical modeling of musical style.

We present a method for hierarchical music clustering, based on compression of strings that represent the music pieces. The method uses no background knowledge about music whatsoever: it is completely general and can, without change, be used in different areas like linguistic classification, literature, and genomics. Indeed, it can be used to simultaneously cluster objects from completely different domains, like with like. It is based on an ideal theory of the information content in individual objects (Kolmogorov complexity), information distance, and a universal similarity metric. The approximation to the universal similarity metric obtained using standard data compressors is called “normalized compression distance (NCD). ” Experiments using our CompLearn software tool show that the method distinguishes between various musical genres and can even cluster pieces by composer. 1.

Modeling long-term dependencies in time series has proved very difficult to achieve with traditional machine learning methods. This problem occurs when considering music data. In this paper, we introduce a model for rhythms based on the distributions of distances between subsequences. A specific implementation of the model when considering Hamming distances over a simple rhythm representation is described. The proposed model consistently outperforms a standard Hidden Markov Model in terms of conditional prediction accuracy on two different music databases. 1.

Computer-based analysis of tonal music has been an active area of research for over a decade. In particular, machine learning based methods have been applied to composing music and solving various problems in musicology such as classification, visualization, search and stylistic

The paper describes a hierarchical Markov modeling strategy that offers the advantages of a statistical approach without constraining the level of analysis. The method can mediate music generation from sample compositions. Some illustrative examples are described. 1.

We are interested in educational software tools that can generate novel jazz solos in a style representative of a body of performed work, such as solos by a specific artist. Our approach is to provide automated learning of a grammar from a corpus of performances. Use of a grammar is robust, in that it can provide generation of solos over novel chord changes, as well as ones used in the learning process. Automation is desired because manual creation of a grammar in a particular playing style is a labor-intensive, trial-and-error, process. Our approach is based on unsupervised learning of a grammar from a corpus of one or more performances, using a combination of clustering and Markov chains. We first define the basic building blocks for contours of typical jazz solos, which we call “slopes”, then show how these slopes may be incorporated into a grammar wherein the notes are chosen according to tonal categories relevant to jazz playing. We show that melodic contours can be accurately portrayed using slopes learned from a corpus. By reducing turn-around time for grammar creation, our method provides new flexibility for experimentation with improvisational styles. Initial experimental results are reported. 1.

submitted for publication Abstract. Modeling long-term dependencies in time series has proved very difficult to achieve with traditional machine learning methods. This problem occurs when considering music data. In this paper, we introduce generative models for melodies. We decompose melodic modeling into two subtasks. We first propose a rhythm model based on the distributions of distances between subsequences. Then, we define a generative model for melodies given chords and rhythms based on modeling sequences of Narmour features. The rhythm model consistently outperforms a standard Hidden Markov Model in terms of conditional prediction accuracy on two different music databases. Using a similar evaluation procedure, the proposed melodic model consistently outperforms an Input/Output Hidden Markov Model. Furthermore, sampling these models given appropriate musical contexts generates realistic melodies. 2 IDIAP–RR 08-51 1

Abstract. Modeling long-term dependencies in time series has proved very difficult to achieve with traditional machine learning methods. This problem occurs when considering music data. In this paper, we introduce a model for rhythms based on the distributions of distances between subsequences. A specific implementation of the model when considering Hamming distances over a simple rhythm representation is described. The proposed model consistently outperforms a standard Hidden Markov Model in terms of conditional prediction accuracy on two different music databases. 2 IDIAP–RR 08-33 1