Download Nonlinear time series analysis of musical signals In this work the techniques of chaotic time series analysis are applied to music. The audio stream from musical recordings are treated as representing experimental data from a dynamical system. Several performance of well-known classical pieces are analysed using recurrence analysis, stationarity measures, information metrics, and other time series based approaches. The benefits of such analysis are reported.
Download Independent subspace analysis using locally linear embedding While Independent Subspace Analysis provides a means of blindly separating sound sources from a single channel signal, it does have a number of problems. In particular the amount of information required for separation of sources varies with the signal. This is as a result of the variance-based nature of Principal Component Analysis, which is used for dimensional reduction in the Independent Subspace Analysis algorithm. In an attempt to overcome this problem the use of a non-variance based dimensional reduction method, Locally Linear Embedding, is proposed. Locally Linear Embedding is a geometry based dimensional reduction technique. The use of this approach is demonstrated by its application to single channel source separation, and its merits discussed.
Download A hierarchical approach to automatic musical genre classification A system for the automatic classification of audio signals according to audio category is presented. The signals are recognized as speech, background noise and one of 13 musical genres. A large number of audio features are evaluated for their suitability in such a classification task, including well-known physical and perceptual features, audio descriptors defined in the MPEG-7 standard, as well as new features proposed in this work. These are selected with regard to their ability to distinguish between a given set of audio types and to their robustness to noise and bandwidth changes. In contrast to previous systems, the feature selection and the classification process itself are carried out in a hierarchical way. This is motivated by the numerous advantages of such a tree-like structure, which include easy expansion capabilities, flexibility in the design of genre-dependent features and the ability to reduce the probability of costly errors. The resulting application is evaluated with respect to classification accuracy and computational costs.
Download A new estimation technique for determining the control parameters of a physical model of a trumpet A new estimation technique is proposed which computes the control parameters of a physical model of a trumpet in order to simulate a recording of a real instrument. First, the physical constraints of the instrument and the prior knowledge about how a player controls a trumpet are described. This is taken into account during the design of the data set and guarantees that these constraints are respected. Then, an estimation procedure minimizes two perceptual similarity criteria in function of the control parameters. The first criterium expresses the difference of the spectral envelopes and the second one the difference in fundamental frequency. An optimization technique is proposed that yields an optimal solution for the fundamental frequency, and a conditional suboptimal solution for the spectral envelope. A robust implementation of the technique was developed for which it is shown that the estimated parameters are unique and that the optimization does not suffer from local minima.
Download On the evaluation of perceptual similarity measures for music Several applications in the field of content-based interaction with music repositories rely on measures which estimate the perceived similarity of music. These applications include automatic genre recognition, playlist generation, and recommender systems. In this paper we study methods to evaluate the performance of such measures. We compare five measures which use only the information extracted from the audio signal and discuss how these measures can be evaluated qualitatively and quantitatively without resorting to large scale listening tests.