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

In this paper we propose a new method for a generalized model representing the time-varying spectral characteristics of quasi harmonic instruments. This approach comprises a linear sourcefilter model, a parameter estimation method and a model evaluation based on the prototype’s variance. The source-filter-model is composed of an excitation source generating sinusoidal parameter trajectories and a modeling resonance filter, whereas basic-splines (B-Splines) are used to model continuous trajectories. To estimate the model parameters we apply a gradient decent method to a training database and the prototype’s variance is being estimated on a test database. Such a model could later be used as a priori knowledge for polyphonic instrument recognition, polyphonic transcription and source separation algorithms as well as for resynthesis.

Download

The automatic segmentation of isolated musical instrument sounds according to the temporal evolution is not a trivial task. It requires a model capable of capturing regions such as the attack, decay, sustain and release accurately for many types of instruments with different modes of excitation. The traditional ADSR amplitude envelope model does not apply universally to acoustic musical instrument sounds with different excitation methods because it uses strictly amplitude information and supposes all sounds manifest the same temporal evolution. We present an automatic segmentation technique based on a more realistic model of the temporal evolution of many types of acoustic musical instruments that incorporates both temporal and spectrotemporal cues. The method allows a robust and more perceptually relevant automatic segmentation of the isolated sounds of many musical instruments that fit the model.

Download

We propose a new statistical model of musical timbre that handles the different segments of the temporal envelope (attack, sustain and release) separately in order to account for their different spectral and temporal behaviors. The model is based on a reduced-dimensionality representation of the spectro-temporal envelope. Temporal coefficients corresponding to the attack and release segments are subjected to explicit trajectory modeling based on a non-stationary Gaussian Process. Coefficients corresponding to the sustain phase are modeled as a multivariate Gaussian. A compound similarity measure associated with the segmental model is proposed and successfully tested in instrument classification experiments. Apart from its use in a statistical framework, the modeling method allows intuitive and informative visualizations of the characteristics of musical timbre.

Download