Catálogo de publicaciones - libros

Image compression is achieved by reducing redundancy between neighboring pixels but preserving features such as edges and contours of the original image. Deterministic and statistical models are usually employed to reduce redundancy. Compression methods that use statistics have heavily been influenced by neuroscience research. In this work, we propose an image compression system based on the concept derived from neural information processing models. The system performance is compared with principal component analysis (PCA) and the discrete cosine transform (DCT) at several compression ratios (CR). Evaluation through both visual inspection and objective measurements showed that the proposed system is more robust to distortions such as ringing and block artifacts than PCA and DCT.

Underdetermined source separation methods often rely on the assumption that the time-frequency source coefficients are independent and Laplacian distributed. In this article, we extend these methods by assuming that these coefficients follow a generalized Gaussian prior with shape parameter . We study mathematical and experimental properties of the resulting complex nonconvex norm optimization problem in a particular case and derive an efficient global optimization algorithm. We show that the best separation performance for three-source stereo convolutive speech mixtures is achieved for small .

In this paper, a new algorithm for source recovery in under-determined Sparse Component Analysis (SCA) or atomic decomposition on over-complete dictionaries is presented in the noisy case. The algorithm is essentially a method for obtaining sufficiently sparse solutions of under-determined systems of linear equations with additive Gaussian noise. The method is based on iterative Expectation-Maximization of a Maximum A Posteriori estimation of sources (EM-MAP) and a new steepest-descent method is introduced for the optimization in the M-step. The solution obtained by the proposed algorithm is compared to the minimum ℓ-norm solution achieved by Linear Programming (LP). It is experimentally shown that the proposed algorithm is about one order of magnitude faster than the interior-point LP method, while providing better accuracy.

Functional Data Analysis (FDA) is used for datasets that are more meaningfully represented in the functional form. Functional principal component analysis, for instance, is used to extract a set of functions of maximum variance that can represent the data. In this paper, a method of Mutual Interdependence Analysis (MIA) is proposed that can extract an equally correlated function with a set of inputs. Formally, the MIA criterion defines the function whose mean variance of correlations with all inputs is minimized. The meaningfulness of the MIA extraction is proven on real data. In a simple text independent speaker verification example, MIA is used to extract a signature function per each speaker, and results in an equal error rate of 2.9 % in the set of 168 speakers.

Existing perceptual models of audio quality, such as PEAQ, were designed to measure audio codec performance and are not well suited to evaluation of audio source separation algorithms. The relationship of many other signal quality measures to human perception is not well established. We collected subjective human assessments of distortions encountered when separating audio sources from mixtures of two to four harmonic sources. We then correlated these assessments to 18 machine-measurable parameters. Results show a strong correlation (r=0.96) between a linear combination of a subset of four of these parameters and mean human assessments. This correlation is stronger than that between human assessments and several measures currently in use.

Successful speech enhancement by convolutive blind source separation (BSS) techniques requires careful design of all aspects of the chosen separation method. The conventional strategy for system initialization in both time- and frequency-domain BSS involves a FIR filter matrix and no data preprocessing; however, this strategy may not be the best for any chosen separation algorithm. In this paper, we experimentally evaluate two different approaches for potentially-improving the performance of time-domain and frequency-domain natural gradient speech separation algorithms – of the signal mixtures, and beamforming initialization for the separation system – to determine which of the two classes of algorithms benefit most from them. Our results indicate that frequency-domain-based natural gradient BSS methods generally need geometric information about the system to obtain any reasonable separation quality. For time-domain natural gradient separation algorithms, either beamforming initialization or prewhitening improves separation performance, particularly for larger-scale problems involving three or more sources and sensors.

For short-time Fourier Transform (STFT) domain ICA, dealing with reverberant sounds is a significant issue. It often invites a dilemma on STFT frame length: frames shorter than reverberation time (short frames) generate incomplete instantaneous mixtures, while too long frames may disturb the separation.

To improve the separation of such reverberant sounds, the authors propose a new framework which accounts for STFT with short frames. In this framework, time domain convolutive mixtures are transformed to STFT domain convolutive mixtures. For separating the mixtures, an approach of applying another STFT is presented so as to treat them as instantaneous mixtures.

The authors experimentally confirmed that this framework outperforms the conventional STFT domain ICA.

The problem of (BSS) of convolved acoustic signals is of great interest for many classes of applications such as in-car speech recognition, hands-free telephony or hearing devices. Due to the convolutive mixing process, the source separation is performed in the frequency domain, using (ICA). However the quality of solution of the ICA-algorithms can be improved by applying . In this paper we present a batch-algorithm for time-frequency masking using the time-frequency structure of separated signals.

In this paper, we investigate the importance of the high frequencies in the problem of convolutive blind source separation (BSS) of speech signals. In particular, we focus on frequency domain blind source separation (FD-BSS), and show that when separation is performed in the low frequency bins only, the recovered signals are similar in quality to those extracted when all frequencies are taken into account. The methods are compared through informal listening tests, as well as using an objective measure.

In this paper, we present a broadband two-microphone blind spatial separation technique by efficiently combining adaptive beamforming (ABF) with multichannel blind deconvolution (MBD). First, the inaccessible source signal streams are partially identified by simple time-delay steering and then are spatially separated through an MBD structure. The proposed spatio-temporal ABF-MBD algorithm exhibits fast convergence properties and high computational efficiency. Numerical experiments illustrate the practical appeal of the proposed method in separating convolutive mixtures of speech within nearly anechoic and also highly reverberant enclosures.