In this paper, we consider a network of sensors in which a fusion center applies a clustering method over the sensor measurements. In order to limit their energy consumption, the sensors transmit their measurements in a compressed form. This paper proposes a novel clustering algorithm that applies directly over compressed data, and that does not require the knowledge of the number of clusters. The proposed algorithm is based on a new cost function for centroid estimation, and a theoretical analysis shows that the cluster centroids are the only minimizers of this cost function. The clustering algorithm then estimates the cluster centroids by looking for the minimizers of the cost function, even when their number is unknown. The proposed algorithm shows performance close to that of the K-means algorithm over compressed data, without need to know the number of clusters.

Random distortion testing (RDT) addresses the problem of testing whether or not a random signal, Ξ, deviates by more than a specified tolerance, τ, from a fixed value, ξ0[1]. The test is nonparametric in the sense that the distribution of the signal under each hypothesis is assumed to be unknown. The signal is observed in independent and identically distributed (i.i.d) additive noise. The need to control the probabilities of false alarm and missed detection while reducing the number of samples required to make a decision leads to the SeqRDT approach. We show that under mild assumptions on the signal, SeqRDT will follow the properties desired by a sequential test. Simulations show that the SeqRDT approach leads to faster decision making compared to its fixed sam-ple counterpart Block-RDT [2] and is robust to model mismatches compared to the Sequential Probability Ratio Test (SPRT) [3] when the actual signal is a distorted version of the assumed signal especially at low Signal-to-Noise Ratios (SNRs).

Signal processing on graphs is a recent research domain that aims at generalizing classical tools in signal processing, in order to analyze signals evolving on complex domains. Such domains are represented by graphs, for which one can compute a particular matrix, called the normalized Laplacian. It was shown that the eigenvalues of this Laplacian correspond to the frequencies of the Fourier domain in classical signal processing. Therefore, the frequency domain is not the same for every support graph. A consequence of this is that there is no non-trivial generalization of Heisenberg's uncertainty principle, that states that a signal cannot be fully localized both in the time domain and in the frequency domain. A way to generalize this principle, introduced by Agaskar and Lu, consists in determining a curve that represents a lower bound on the compromise between precision in the graph domain and precision in the spectral domain. The aim of this paper is to propose a characterization of the signals achieving this curve, for a larger class of graphs than the one studied by Agaskar and Lu.

Random distortion testing (RDT) introduced in [1] is aimed at detecting any significantly big distortion of a signal with respect to a model of this signal, in presence of noise and without prior knowlegde on the distortion distribution. The RDT formulation makes it possible to state the standard change-in-mean detection problem differently. It leads to the Block-RDT approach that requires no iid assumption and no prior knowledge on the distributions of the observations before and after change. The optimal tests derived in the Block-RDT approach are alternative to Shewhart charts and outperform the latter by accounting for possible model mismatches that cause likelihood theory to fail. Experimental results dedicated to the detection of steps in a random process illustrate our intention.

We propose a novel method for noise power spectrum estimation in speech enhancement. This method called extended-DATE (E-DATE) extends the d-dimensional amplitude trimmed estimator (DATE), originally introduced for additive white gaussian noise power spectrum estimation in “Robust estimation of noise standard deviation in presence of signals with unknown distributions and occurrences” (D. Pastor and F. Socheleau, IEEE Trans. Signal Processing, vol. 60, no. 4, pp. 1545-1555, Apr. 2012) to the more challenging scenario of non-stationary noise. The key idea is that, in each frequency bin and within a sufficiently short time period, the noise instantaneous power spectrum can be considered as approximately constant and estimated as the variance of a complex gaussian noise process possibly observed in the presence of the signal of interest. The proposed method relies on the fact that the Short-Time Fourier Transform (STFT) of noisy speech signals is sparse in the sense that transformed speech signals can be represented by a relatively small number of coefficients with large amplitudes in the time-frequency domain. The E-DATE estimator is robust in that it does not require prior information about the signal probability distribution except for the weak-sparseness property. In comparison to other state-of-the-art methods, the E-DATE is found to require the smallest number of parameters (only two). The performance of the proposed estimator has been evaluated in combination with noise reduction and compared to alternative methods. This evaluation involves objective as well as pseudo-subjective criteria.

We consider the problem of estimating a deterministic finite alphabet vector f from underdetermined measurements y = A f , where A is a given (random) n × N matrix. Two new convex optimization methods are introduced for the recovery of finite alphabet signals via ℓ1-norm minimization. The first method is based on regularization. In the second approach, the problem is formulated as the recovery of sparse signals after a suitable sparse transform. The regularization-based method is less complex than the transform-based one. When the alphabet size p equals 2 and (n, N) grows proportionally, the conditions under which the signal will be recovered with high probability are the same for the two methods. When p > 2, the behavior of the transform-based method is established. Experimental results support this theoretical result and show that the transform method outperforms the regularization-based one.

We consider the problem of testing whether the energy of a random signal projected onto a known subspace exceeds some specified value$\tau \geq 0$. The probability distribution of the signal is assumed to be unknown and this signal is observed in additive and independent white Gaussian noise with known variance. The proposed theoretical framework relies on the invariance of the problem and the resulting test is shown to be uniformly most powerful invariant in an extended sense suitable for random signals. This work extends Scharf and Friedlander’s matched subspace detector.

We address Random Distortion Testing (RDT), that is, the problem of testing whether the Mahalanobis distance between a random signal Θ and a known deterministic model θ0exceeds some given τ ≥ 0 or not, when Θ has unknown probability distribution and is observed in additive independent Gaussian noise with positive definite covariance matrix. A suitable optimality criterion for RDT is presented and theoretical results on optimal tests for this criterion are given. Several applications of these results are presented and analyzed. They address the detection of signals in case of model mismatch and the detection of deviations from model θ0.

Statistical properties of real-valued symmetric -stable noise after short-time Fourier transformation are derived. Circularity, stationarity and dependence between the real and imaginary components are studied as a function of the STFT parameters and the stability index .

This paper introduces a method to automatically detect AutoPEEP (pulmonary distension), a frequent asynchrony in the patient-ventilator interface. The detection algorithm is developed based on a robust non-parametric hypothesis testing that requires no prior information on the distribution of the signal. The experiment results have shown that the proposed algorithm provide relevant AutoPEEP detection on both simulated and real data.

A new estimator, optimal in the frequency domain with respect to the masking properties of the human auditory system, is proposed. This new filtering technique prevents the emergence of post-filtering isolated tonals that increase the musical noise perception. Experimental results by means of objective tests show that this technique improves the enhanced speech quality.

We address the problem of blind source separation in the underdetermined and instantaneous mixture case. The proposed method is based on an algorithm developed by Aissa-El-Bey and al.. This algorithm requires a good choice of the noise threshold and does not take into account the noise contribution in the inversion process. In order to overcome these drawbacks, this paper presents a robust underdetermined blind source separation approach. Robustness is achieved by estimating the noise standard deviation and using this estimate in the inversion process and the expression of the noise threshold. The good performance of the proposed method is shown by comparison with state-of-the-art methods.

Monitoring sea surface temperature (SST) is of high interest. The dynamics of climatic events such as tropical storms or cyclones are closely related to the sea surface temperature. An important research topic in oceanography is about the understanding of the interactions between the ocean's surface and it's deeper layers. In Lapeyre et al. show that in the case of baroclinic unstable flows, the potential vorticity of mesoscale and submesoscale structures in the ocean interior are strongly correlated to the surface density structures. Thus, in frontal regions, knowledge of the ocean surface structures regularity gives insights about the 3D dynamics of the ocean. In this paper we focus on the analysis of the SST images level lines regularity from satellites images in regions in the neighborhood of oceanic fronts. Two regions of interest are considered: the region of Benguela and the region of Malvinas. Our investigations suggest that SST fronts belong to the class of statistical self similar curves. Taking into account self similar curves properties, we propose local curvature based descriptors for SST front region regularity characterization. We experimentally assess the efficiency of our descriptors by measuring their ability to capture SST image regularity seasonal variations.

In this paper, we investigate the recognition of cochlear implantlike spectrally reduced speech (SRS) using conventional speech features (MFCCs and PLP coefficients) and HMM-based ASR. The SRS was synthesized from subband temporal envelopes extracted from original clean speech for testing, whereas the acoustic models were trained on a different set of original clean speech signals of the same speech database. It was shown that changing the bandwidth of the subband temporal envelopes had no significant effect on the ASR word accuracy. In addition, increasing the number of frequency subbands of the SRS from 4 to 16 improved significantly the system performance. Furthermore, the ASR word accuracy attained with the original clean speech, by using both MFCC-based and PLP-based speech features, can be achieved by using the 16-, 24-, or 32-subband SRS. The experiments were carried out by using the TI-digits speech database and the HTK speech recognition toolkit.

This correspondence investigates the recognition of cochlear implant-like spectrally reduced speech (SRS) using mel frequency cepstral coefficient (MFCC) and hidden Markov model (HMM)-based automatic speech recognition (ASR). The SRS was synthesized from subband temporal envelopes extracted from original clean test speech, whereas the acoustic models were trained on a different set of original clean speech signals of the same speech database. It was shown that changing the bandwidth of the subband temporal envelopes had no significant effect on the ASR word accuracy. In addition, increasing the number of frequency subbands of the SRS from 4 to 16 improved significantly the system performance. Furthermore, the ASR word accuracy attained with the original clean speech can be achieved by using the 16-, 24-, or 32-subband SRS. The experiments were carried out by using the TI-digits speech database and the HTK speech recognition toolkit.

This paper provides asymptotic properties of the autocorrelation functions of the wavelet packet coefficients of a fractional Brownian motion. It also discusses the convergence speed to the limit autocorrelation function, when the input random process is either a fractional Brownian motion or a wide-sense stationary second-order random process. The analysis concerns some families of wavelet paraunitary filters that converge almost everywhere to the Shannon paraunitary filters. From this analysis, we derive wavelet packet based spectrum estimation for fractional Brownian motions and wide-sense stationary random processes. Experimental tests show good results for estimating the spectrum of1/fprocesses.

We present two new noise variance estimation methods for OFDMA signals transmitted through an unknown multipath fading channel. We focus on blind estimation as it does not require any pilot sequences and is therefore applicable to contexts, such as cognitive radio for instance, where little prior signal knowledge is available. The two estimators are respectively based on the time-frequency sparsity of OFDMA signals and on the redundancy induced by the cyclic prefix. Numerical simulations compare the performance of the two algorithms and highlight their complementarity.

The paper presents a general framework for change detection in radar images, for an operational purpose and in the context of environmental monitoring. This framework is based on a processing which provides highly sparsifiable representations of data. This processing is called turbo-median and is a combination of the sample median robustness and the turbo principle for iteratively correcting errors. The turbo-median processing of a scene is an homogenized representation based on an iterative median and which consists in spreading the statistically more robust measurements of the scene under consideration over the size of the image representing this scene. It allows for reducing the change detection problem into the problem of detecting a signal, with unknown distribution, in additive noise.

This paper deals with musical noise resulting from subtractive type algorithms and especially Wiener filtering. We compare several methods that introduce perceptually motivated modifications of standard Wiener filtering and we propound a new speech enhancement technique. It aims to improve the quality of the enhanced speech signal provided by standard Wiener filtering by controlling the latter via a second filter regarded as a psychoacoustically motivated weighting factor. According to objective measures, the described process results in significant reduction of musical noise.

This paper presents a new sigmoid-based wave shrink function. The shrinkage obtained via this function is particularly suitable to reduce noise without impacting significantly the statistical properties of the signal to be recovered. The proposed WaveShrink function depends on a parameter that makes it possible to control the attenuation degree imposed to the data, and thus, allows for a flexible shrinkage.

This paper proposes a new structure for a scalable codec. Our proposed codec works with 10 ms input frame for wideband speech and audio signals at bit rates ranging from 8 to 32 kbit/s. The core layer is the ITU-T G.729 at 8 kbit/s producing a narrowband output. The first enhancement layer is a band-width extension providing a wideband output with 2 kbit/s. The second enhancement layer is based on algebraic quantization of wavelet packet coefficients and improves gradually the synthesized signal as the bitrate increases. For speech signals, at bitrates of 24 and 32 kbit/s, the codec is shown to be equivalent to the ITU-T G.722 codec at 56 and 64 kbit/s, respectively. Moreover, the codec at 32 kbit/s is assessed to be equivalent to the recently standardized embedded codec ITU-T G.729.1 at the same bitrate with a lower algorithmic delay.

This paper proposes a new quantization for transform coefficients based on algebraic quantization. The coefficients are represented by a few pulses multiplied by a unique amplitude. The coefficients to be transmitted are selected by optimizing an error criterion, that determines the signs, positions and amplitudes of the pulses. This simple quantization has been implemented in a wavelet-based wideband scalable coder, and has been proved to provide a perceptually better quality than SPIHT on speech signal and music.

This paper presents a scalable wideband speech codec working at bitrates ranging from 8 to 32 kbit/s. The core layer is the ITU-T G. 729 at 8 kbit/s. A first enhancement layer is a bandwidth extension algorithm requiring 2 kbit/s to widen the G. 729 narrow band output speech. The difference between the wideband original and reconstructed signal is transformed in the time-frequency domain by a full wavelet decomposition. The resulting coefficients are quantized by an embedded quantizer at 22 kbit/s. Listening tests show the relevance of such a scheme when compared to a pure wavelet packet decomposition. In addition, listening tests suggest that the proposed codec is equivalent to the ITU-T G. 722 at 48 kbit/s for speech signals.

Consider n-dimensional observations where random signals are present or absent in independent and additive white Gaussian noise (AWGN) with standard deviation sigmaO. On the basis of recent results in statistical decision theory, this paper presents a new algorithm for estimating sigmaOwhen the signals are less present than absent and have unknown probability distributions. The bias, the consistency and the minimum attainable mean square estimation error of the estimator we propose are still unknown. However, experimental results are very promising. When the minimum-probability-of-error decision scheme for the non-coherent detection of modulated sinusoidal carriers in independent AWGN is tuned with the estimate instead of the true value sigmaO, the binary error rate obtained tends rapidly to the optimal error probability after a few hundred observations

This paper introduces a new approach for the automatic segmentation of corpora dedicated to speech synthesis.The main idea behind this approach is to merge the outputs of three segmentation algorithms.The first one is the standard HMM-based (Hidden Markov Model) approach.The second algorithm uses a phone boundary model, namely a GMM (Gaussian Mixture Model).The third method is based on Brandt's GLR (Generalized Likelihood Ratio) and aims to detect signal discontinuities in the vicinity of the HMM boundaries.Different fusion strategies are considered for each phonetic class.The experiments presented in this paper show that the proposed approach yields better accuracy than existing methods.

A new sharp upper bound for the probability of error of the likelihood ratio test is given for the detection in white Gaussian noise of any random vector whose norm is greater than, or equal to, a given value and whose probability of presence is less than, or equal to, one half. Also, a new test for the detection of such vectors is described. This test does not depend on the distribution of the signal vector but nevertheless its probability of error is less than, or equal to, the given upper bound.