The separation of musical instruments acoustically mixed in one source is a very active field which has been approached from many different viewpoints. This article compares the blind source separation perspective and oscillatory correlation theory taking the auditory scene analysis as the point of departure (ASA). The former technique deals with the separation of a particular signal from a mixture with many others from a statistical point of view. Through the standard Independent Component Analysis (ICA), a blind source separation can be done using the particular and the mixed signals' statistical properties. Thus, the technique is general and does not use previous knowledge about musical instruments. In the second approach, an ASA extension is studied with a dynamic neural model which is able to separate the different musical instruments taking a priori unknown perceptual elements as a point of departure. Applying an inverse transformation to the output of the model, the different contributions to the mixture can be recovered again in the time domain.

Download

This paper describes the first version of a Java archive (a term basically equivalent to ‘library’ in other programming languages) that has been developed and made available as public domain software for the benefit of the DAFX community, and the COSTG6 web pages in particular. The library is available both as source code and ready to run bytecodes. The archive defines an easy to use set of classes that are modelled after an effects processor. Ready made classes like ‘Effect’, ‘Page of Parameters’, ‘Integer Range Parameter’, ‘Real Range Parameter’, etc. serve as a basis to implement effects and share them. Effects can run from web pages or as stand alone applications, sharing unified look and feel in a platform independent graphical user interface. The programmer only needs to specify the parameters the effect will use, and the method (function) that will apply the effect to each new sample. An automatic GUI interface is created, that enables the adjustment of parameters as well as the specification of input and output files to be used during processing. Developing Java audio effects according to the proposed scheme will allow transparent integration into more complex multiband and multieffect architectures that will be added on a second version of the archive.

Download

This paper proposes a new scheme to improve the source separation problem aimed to microphone array applications like WFS based teleconference systems. A multichannel, sub-band approach to reduce computational complexity is presented. Also, instead of using the LMS adaptive algorithm, a new system based on hybrid Conjugate Gradient-nLMS techniques is developed to accelerate the convergence time. This adaptive algorithm is controlled by a voice activity detector block that basically detects double talk situations and freezes the adaptation process to avoid the appearance of sound artifacts which may cause a significant degradation of the recovered signals and have a great impact in the quality of the full system.

Download

Conjugate Gradient (CG) techniques are suitable for resolution of time-variant system identification problems: adaptive equalization, echo cancellation, active noise cancellation, linear prediction, etc. These systems can be seen as optimization problems and CG techniques can be used to solve them. It has been demonstrated that, in the single-channel case, the conjugate gradient techniques provide a similar solution in terms of convergence rate than those provided by the recursive least square (RLS) method, involving higher complexity than the least mean square (LMS) but lower than RLS without stability issues. The advantages of these techniques are especially valuable in the case of high complexity and magnitude problems like multi-channel systems. This work develops CG algorithm for the adaptive MIMO (multiple-input and multiple-output) systems and tests it by solving a multichannel acoustic echo cancellation (MAEC) problem.

Download