Catálogo de publicaciones - libros

This paper presents a new algorithm for solving the permutation ambiguity in convolutive blind source separation. When transformed to the frequency domain, the source separation problem reduces to independent instantaneous separation in each frequency bin, which can be efficiently solved by existing algorithms. But this independency leads to the problem of correct alignment of these single bins which is still not entirely solved. The algorithm proposed in this paper models the frequency-domain separated signals using the generalized Gaussian distribution and utilizes the small deviation of the exponent between neighboring bins for the detection of correct permutations.

Discovering a representation that allows auditory data to be parsimoniously represented is useful for many machine learning and signal processing tasks. Such a representation can be constructed by Non-negative Matrix Factorisation (NMF), which is a method for finding parts-based representations of non-negative data. Here, we present a convolutive NMF algorithm that includes a sparseness constraint on the activations and has multiplicative updates. In combination with a spectral magnitude transform of speech, this method extracts speech phones that exhibit sparse activation patterns, which we use in a supervised separation scheme for monophonic mixtures.

This paper proposes a way to implement a time-domain blind separation algorithm for convolutive mixtures of source signals. The approach provides another form of the algorithm by discrete Fourier transform and has the possibility of designing a separating filter in the frequency domain, without bothering about the permutation problem inherent in frequency-domain blind separation approach. This paper also shows a technique to improve separation performance in the frequency domain. The validity of our approach was demonstrated by performing an experiment on separation for convolutive mixtures of two speeches.

Non-negative spectrogram factorization has been proposed for single-channel source separation tasks. These methods operate on the magnitude or power spectrogram of the input mixture and estimate the magnitude or power spectrogram of source components. The usual assumption is that the mixture spectrogram is well approximated by the sum of source components. However, this relationship additionally depends on the unknown phase of the sources. Using a probabilistic representation of phase, we derive a cost function that incorporates this uncertainty. We compare this cost function against four standard approaches for a variety of spectrogram sizes, numbers of components, and component distributions. This phase-aware cost function reduces the estimation error but is more affected by detection errors.

Auditory scene analysis is extremely challenging. One approach, perhaps that adopted by the brain, is to shape useful representations of sounds on prior knowledge about their statistical structure. For example, sounds with harmonic sections are common and so time-frequency representations are efficient. Most current representations concentrate on the shorter components. Here, we propose representations for structures on longer time-scales, like the phonemes and sentences of speech. We decompose a sound into a product of processes, each with its own characteristic time-scale. This demodulation cascade relates to classical amplitude demodulation, but traditional algorithms fail to realise the representation fully. A new approach, probabilistic amplitude demodulation, is shown to out-perform the established methods, and to easily extend to representation of a full demodulation cascade.

This article provides an overview of the first stereo audio source separation evaluation campaign, organized by the authors. Fifteen underdetermined stereo source separation algorithms have been applied to various audio data, including instantaneous, convolutive and real mixtures of speech or music sources. The data and the algorithms are presented and the estimated source signals are compared to reference signals using several objective performance criteria.

This paper addresses the tracking capability of blind source separation algorithms for rapidly time-varying sensor or source positions. Based on a known algorithm for blind source separation, which also allows for simultaneous localization of multiple active sources in reverberant environments, the source separation performance will be investigated for abrupt microphone array rotations representing the . After illustrating the deficiencies in source-tracking with the given efficient implementation of the BSS algorithm, a method to ensure robust source separation even with abrupt microphone array rotations is proposed. Experimental results illustrate the efficiency of the proposed concept.

Recently Single Channel ICA has been proposed where it can be shown that the algorithms learn temporal filters for separating the different components. Here we consider the natural extension to learning a set of space-time separating filters. We argue that these are capable of separation above and beyond that possible using only spatial or temporal methods alone. We then consider the potential of these ideas when applied to Ictal Electroencephalographic (EEG) data and Brain Computer Interaction (BCI).

The detection of the basic electric rhythm (BER), composed of 3 cycles/minute oscillation, can be performed using SQUID sensors. However the electric response activity (ERA), which is generated when the stomach is performing a mechanical activity, was detected mainly by invasive electrical measurements and only recently one report was published dealing with its detection by magnetic measurements. This study was performed with the aim to detect and extract the ERA and ECA noninvasively before and after a meal. After acquire MGG recordings the signals were processed to extract both source components and remove cardiac interference and others interferences by an algorithm based on Dependent Component Analysis (DCA) then autoregressive and wavelet analysis was performed. Therefore, first, we can compare their relative amplitudes in the time or frequency domain, and get evidences of ERA signal. Second, we can get the spatial contribution from each channel to the source signal extracted. Finally, results have shown that there is an increase in the signal power at higher frequencies around (0.6-1.3 Hz) from ERA source component usually associated with the basic electric rhythm (ECA source component). We show that the method is effective in removing interference signals of MGG recordings, and is computationally efficient.

The continuous demand for high performance and low cost electrocardiogram (ECG) processing systems have required the elaboration of more and more efficient and reliable ECG compression techniques. Such techniques face a tradeoff between compression ratio and retrieved quality, where the decrease of the last can compromise the subsequent use of the signal for clinical purposes. The objective of this work is to evaluate the validity and performance of an independent component analysis (ICA) based scheme used to efficiently compress ECG signals while introducing tests for a different type of record of the electrical activity of the heart, such as fetal magnetocardiogram (fMCG). As a result, the reconstructed signals underwent negligible visual deterioration, while achieving promising compression ratios.