SEPARATION OF ACOUSTIC SIGNALS USING SELF-ORGANIZING NEURAL NETWORKS. Temujin Gautama & Marc M. Van Hulle
|
|
- Brenda Burke
- 6 years ago
- Views:
Transcription
1 SEPARATION OF ACOUSTIC SIGNALS USING SELF-ORGANIZING NEURAL NETWORKS Temujin Gautama & Marc M. Van Hulle K.U.Leuven, Laboratorium voor Neuro- en Psychofysiologie Campus Gasthuisberg, Herestraat 49, B-3000 Leuven, BELGIUM Tel.: , Fax: ftemu,marcg@neuro.kuleuven.ac.be Abstract Spectral modeling is an essential component in many signal processing applications, such as speech enhancement and sound monitoring. This paper will demonstrate its use in the separation of acoustic sources from a compound signal that is registered by one sensor. Our technique distinguishes itself from the popular blind source separation procedure by its much higher noise insensitivity and its ability to cope with varying as well as non-square mixing conditions. INTRODUCTION In the neural network literature, the separation of acoustic sources is mostly described in the context of blind source separation (BSS) (for references, see [4]). The source signals are assumed to be statistically independent and they are separated from the recorded mixture signals without knowing the mixture process. This process is generally assumed to be linear, with as many sensors as there are sources, and with a mixing matrix that is both invertible and constant over time. Under these conditions, the mixing process can be modeled and the source signals can be recovered. However, since the BSS algorithms are essentially based on amplitude information, they tend to be quite sensitive to noise [1]. If there is only a single sensor present, BSS of two sources becomes a nonsquare problem, and the current BSS algorithms cannot be used. One way to tackle this problem is to model the spectral characteristics of the acoustic sources. Single microphone speech enhancement [7] and sound monitoring [6] are well-known applications in this eld, where acoustic sources are rst modeled using spectral information, after which these models are applied either to cancel out noise, to detect changes in the acoustic sources, or to identify the presence of an unknown source. Modeling the sources using their spectral characteristics is expected to yield an increased resistance to noise and to variabilities in the mixing pro-
2 Source 1 Source 2 S 1 Compound S 2 C I 1 I 2 M 1 M 2 φ φ O 1 O 2 Output 1 Output 2 Figure 1: Diagram of the system. All signals are represented by their complex Fourier spectrum. cess. These reasons form the main motivations to adopt a spectral modeling approach to source separation. The models considered in this paper will be developed using self-organizing learning rules trained on spectral information of music signals. PROCEDURE Basic Strategy Figure 1 shows the overall system setup. The source signals S 1 and S 2 are linearly mixed so as to produce the compound signal C (note that all signals are represented by their complex short-time Fourier transforms (STFT) S, instead of by their amplitude spectra S). The compound signal C is applied to the two branches in the system, one for each source signal. In every branch, the estimated spectrum of the other source is subtracted from C and the result, I n, is quantized by the model M n, (n 2 f1; 2g). The quantized signal is then phase aligned () to the model's input signal I n (dashed line), so as to produce the source signal estimate O n after a number of iterations. The exact nature of both the quantization and the phase alignment will be elaborated upon in sections 2.2 and 2.3. The quantization models will be trained using self-organizing learning rules, as will be explained in section 2.3. As an example, we will consider signals taken from two musical instruments (oboe and piano). Quantization The spectral model of each instrument, M n, consists of a set of codebook vectors fq n;i g each of which represents a \prototypical" Fourier spectrum. These vectors should be translation (in time) invariant so as to avoid redundancy in the representation. Hence, M n will consist of amplitude spectra Q n;i, rather than complex Fourier spectra Q n;i. At time t, the winning codebook Q n;c for input (amplitude) spectrum I n will be selected using the dot
3 product criterion, i.e. jq n;c I n j jq n;i I n j; 8 Q n;i 2 M n : (1) The reconstructed amplitude spectrum O n is dened as the projection of I n onto the winner Q n;c and represents the estimated amplitude spectrum of one source signal. Next, O n has to be supplemented with a phase spectrum. If not, transformation to the time domain would be impossible, since the phase spectrum contains information about the harmonic structure of the signal. The matching process (eq. 1), however, should not be performed on complex codebook vectors, since these are not translation invariant. Hence, we have opted for a two-step procedure: rst the matching is performed on the amplitude spectrum only (Amplitude Quantization), after which the reconstruction is combined with the corresponding phase spectrum (Complex Quantization and Phase Alignment). This two-step procedure and the training of the models will be explained in section 2.3. Building and Applying the Quantization Models Amplitude Quantization Model. The codebook for model M n, fq n;i g, is determined, for every instrument separately, using self-organizing learning rules. We will compare three such rules: kmer [5], Kohonen's SOM [2] and the k-means algorithm [3]. In the kmer (kernel-based Maximum Entropy Rule, [5]), every neuron i is determined by its traditional weight vector w i, supplemented by a radius i. A given neuron is active if the Euclidean distance between its weight vector and the input pattern is smaller than the neuron's radius. The rule is aimed at building a topographic map in which all neurons have the same activation probability, the exact level of which is determined by the parameter. This is achieved by adjusting both the weight vectors and their radii in an iterative manner (note that the activation regions are allowed to overlap). The update rule incorporates a neighborhood function (in lattice space) whose width decreases over time in order to yield topographic maps. Training patterns for the learning rules consist of amplitude spectra coming from a single instrument. They are thresholded in energy in order to avoid training on irrelevant spectra by using the following criterion: input pattern will be considered for training when: E < E > training set ; (2) 4 where E is the signal energy of input pattern. Since the actual quantization will be performed by means of projections, all patterns are normalized prior to training. The result of this stage is a quantized amplitude spectrum for each instrument (computed as the projection of I n onto the winner Q n;c ). We now have to extend this representation with a phase spectrum in order to restore the harmonic structure of the time signal.
4 Complex Quantization Model. The prototype amplitude spectra Q n;i will be extended with representative phase spectra so as to preserve the phase relationships between frequency components. This will be done only once, after training, by combining every prototype spectrum Q n;i with the phase spectrum of its closest matching input spectrum I : jq n;i I j ki k jq n;i I j ki ; 8: (3) k The result of this stage is a reconstruction of the complex spectrum. However, the exact positioning in the time slice still has to be determined. Phase Compensation. The reconstructed, complex spectrum O n is the Fourier transform of a time slice o n (t). If we assume the phase relations between the frequency components to be correct, then o n (t) represents a time slice that corresponds to i n (t), except for a time shift within the time slice, t (the phase spectrum determines both the phase relations between components and the exact positioning in time). This t can be computed as the period of time over which o n (t) has to be circularly shifted over the time slice, so as to optimally correlate with input signal i n (t) (circularly shifted, since the discrete Fourier transform assumes the signals to be periodic over the given time slice): t = arg max[ji n (t) o n (t)j] = arg max[jf?1 fi n O ngj]; (4) where denotes the correlation function, F?1 the inverse Fourier transform, I n is the Fourier transform of i n (t) and O n is the complex conjugate of the Fourier transform of o n (t). As a result, these three stages yield time signals that approximate the original source signals in frequency content, phase relations and positioning in time. Source Separation Once the spectral models have been trained, the separation of the acoustic sources can be performed by means of a process that iteratively renes the spectral estimates of the sources at every time step t. Figure 1 shows that the input to quantization model M 1 is formed by the subtraction of the estimated spectrum of source 2 at iteration (k? 1) from the compound spectrum C(t), and vice versa. To improve performance, a competitive element has been included: at iteration k, the model M m (m 6= n) that nds the worse match to its input, will quantize (C(t)? O k n(t)), rather than (C(t)? O (k?1) n (t)). The separation will now be explained at iteration k. 1. The inputs to the quantization models, I n (t), are set to (C? O (k?1) m ), with m 6= n.
5 2. Amplitude quantization. The closest amplitude matches Q n;c are found for every model M n using eq. (1). The cosines of the angles between Q n;c and the input amplitude spectrum I n (t) are determined. M win denotes the model that yields the highest cosine. I win (t) is projected onto the model's best matching vector Q win;c. The length of this projection corresponds to a gain factor that is applied to Q win;c. 3. Complex quantization. Q win;c is combined with its corresponding phase spectrum. 4. Phase alignment on this complex spectrum yields O k win, using eq. (4). 5. The complex residue spectrum (C(t)? O k win (t)) is quantized by the remaining model M m and phase aligned resulting in O k m(t) (m 6= win). At iteration 0, the inputs to the quantization models M n are all set to C (i.e. the initial spectral estimates of the models are set to the null vector). The number of iterations is set to 3 in all simulations described in this paper. When the competitive element was not included, the performance decreased by 0:2 db and more than 3 iterations did not lead to better results. SIMULATIONS Signal Generation and Evaluation The signals that were used in this section have been generated on a Crystal 4232 audio controller (Yamaha OPL3 FM synthesizer) at a sampling rate of Hz, using the simulated sounds of an oboe and a piano. The compound signal was created by linearly mixing two separate time signals (9 seconds of music): c(t) = 0:5 s1(t) + 0:5 s2(t). The STFT is computed every 256 samples, resulting in a 3/4 overlap; the length of one time window is 1024 samples. At this stage, no windowing is applied so as not to interfere with the correlation strategy of the phase alignment (eq. 4). For the inverse transform, the consecutive complex spectra are transformed to the time domain. Every time window is then Hamming windowed and oset in time (spectrum i will have an oset of (256 i) samples), after which they are summed so as to yield the reconstructed time signal, o n (t). The two training sets consist of the short-time Fourier spectra of time signals of two octaves (15 notes, 10 seconds) played on the respective instruments (the same octaves for both oboe and piano). The resulting spectra are then thresholded using eq. 2 and normalized, resulting in training sets of 364 and 387 complex spectra for oboe and piano respectively. The compound signal consists only of notes included in the training sets. In all simulations, the signal-to-noise ratio (SNR) has been used to evaluate performances. It is dened for the original time signal s n (t) and its
6 reconstruction o n (t) as follows: X T?1 1 SN R(s n ; o n )[db] =?10 log 10 ( s n(t) T t=0 sn? o n(t) on ) 2 ; (5) where sn and on are the standard deviations of s n and o n, and T is the length of the shorter signal. The SNR of a round transformation (STFT, followed by its inverse) is db for the oboe and db for the piano training signal. Dimensionality Reduction In order to facilitate and speed up the training process of the quantization models, a dimensionality reduction has been performed on the amplitude spectra of both training signals separately, by means of the Karhunen-Loeve transform. First, the principal components of the amplitude spectra (512 dimensions) are computed, after which every data point is projected onto the rst N ( 512) principal components. To monitor the quality of the reconstruction of the training signals as a function of N, the amplitude spectra coming from the STFT were rst reduced in dimensionality, after which they were reconstructed, extended with their original phase spectra and, nally, transformed back to the time domain. Figure 2A shows the quality of the reconstructed time signals as a function of N. The piano model reaches its maximum for N 300 and the oboe model for N 100. The dierence in performance between instruments corresponds to the quality of the above mentioned round transformation (14.88 db and db for oboe and piano) and hence is not due to the Karhunen- Loeve transform. To monitor the inuence of the dimensionality reduction on the overall separation performance, all learning rules were trained for ve dierent values of N (20, 40, 100, 200 and 300). Training the Models The goal of training is the generation of a codebook of 49 vectors (kmer and SOM are congured as 7x7 lattices). During all training sessions, the mean squared error on the training set is monitored, so as to nd good parameter sets for the dierent learning rules (learning rates, length of training and, for the kmer and the SOM algorithm, the neighborhood cooling schemes). Both the kmer and the SOM algorithm are initialized with the same random congurations, while k-means is initialized with vectors randomly taken from the training set. Figure 2B shows the SNRs of the dierent models' reconstructed piano training signals, evaluated in function of the dimensionality reduction (oboe results not shown). The solid lines show the quality of the reconstruction when the original phase spectrum is used to reconstruct the complex Fourier spectrum (i.e. Amplitude Quantization only). Dashed lines represent
7 A 22.0 B kmer Quantization SOM Quantization Oboe k-means Quantization 1 Piano SNR [db] SNR [db] Number of Principal Components Number of Principal Components 320 Figure 2: Performance (S/N in db) in function of the number of principal components of A) the reconstruction of the training signals after dimensionality reduction (the phase is untouched); B) the reconstruction of the piano training signal after amplitude quantization (solid line) and complex quantization (dashed line) for the dierent learning rules. the SNRs of the reconstruction where the Complex Quantization and Phase Alignment of section 2.3 have been used. Figure 2B also demonstrates the robustness of the dierent learning rules with respect to the dimensionality of the training data. The k-means performance is prone to the sparseness that is due to a higher dimensionality. Thus, it cannot take advantage of an increasing number of eigenvectors which allows for better spectral reconstructions, whereas the kmer and the SOM can (cfr. Fig. 2A). Separation The complete system is tested and evaluated for the dierent models (ve dierent sets for all three learning rules: N = 20; 40; 100; 200; 300). Figure 3A shows the performances for the dierent models, evaluated in function of the number of principal components. Again, kmer and SOM's performances do not degrade with increasing dimensionality. In order to better evaluate the performances, we considered two more models: M tr and M sep. M tr corresponds to a \perfect" quantization of the training set: the models consist of all normalized spectra of the training sets. M sep corresponds to optimal codebooks with respect to the separation: they consist of all normalized spectra of the separate instruments in the compound signal. The results are M tr (7.97 db and db) and M sep (9.07 db and db). CONCLUDING REMARKS In this paper, spectral models have been applied in the context of the separation of acoustic sources. The models themselves have been developed
8 SNR [db] 10.0 A kmer SOM k-means Number of Principal Components B SNR Separation [db] Oboe Piano SNR Input [db] 16.0 Figure 3: A) Performances of the system for the dierent models, evaluated in function of the dimensionality reduction. Solid lines represent the performances for the piano, dashed lines the ones for the oboe. B) Performance of the system in a noisy environment using self-organizing learning rules. The recognition/separation is based on the projection of the compound signal onto the models' prototypical spectra. It is due to this projection that our models are invariant to noise: when uniform noise is added to the compound signal (Fig. 3B), the performance is still reasonable (5.59 db) for a SNR of 4.7 db of the compound signal (this corresponds to a noise amplitude of 30%). In applications in the eld of sound monitoring, these models would be able to recognize the expected acoustic events, even in a noisy environment. The complete system could also be used to detect and separate unknown events from the modeled ones and even perform on-line modeling. Contrary to blind source separation, our system does not assume a xed mixing. Simulations have been performed where the mixing was variable. The mixing coecients for the two instruments varied as a sine and a cosine, both with an oset of 1.1, with one cycle over the total signal length. Such a variable mixing could be caused by moving acoustic sources. When using the SOM models trained on 100 principal components, the separation performance hardly degraded and stayed at 5.83 db for the oboe and 9.78 db for the piano signal, compared to 6.17 db and 9.75 db when the mixing is constant (Fig. 3A). If two sensors were present, the sources could easily be located and tracked through time, since the mixing coecients can be determined. All these are topics for further investigation. ACKNOWLEDGMENTS T.G. is supported by a scholarship from the Flemish Institute for the promotion of Scientic Technological Research in Industry (I.W.T.). M.M.V.H. is a research associate of the Fund for Scientic Research { Flanders (Belgium)
9 and is supported by research grants received from the Fund for Scientic Research (G ), the National Lottery (Belgium) ( ), the Flemish Ministry of Education (GOA 95/99-06), and the Flemish Ministry for Science and Technology (VIS/98/012). REFERENCES [1] Herrmann, M., and Yang, H.H. (1996). Perspectives and limitations of self-organizing maps in blind separation of source signals. In S.-I. Amari, L. Xu, L.-W. Chan, I. King, and K.-S. Leung (Eds.), Progress in Neural Information Processing (Vol. 2, pp ). Proceedings of the International Conference on Neural Information Processing (ICONIP'96), London: Springer-Verlag. [2] Kohonen, T. (1982). Self-organized formation of topologically correct feature maps. Biol. Cybern., 43, pp [3] Krishnaiah, P.R., and Kanal, L.N. (1982). Classication, Pattern Recognition, and Reduction of Dimensionality, Handbook of Statistics, vol. 2, Amsterdam: North Holland. [4] Van Hulle, M.M. (1998). Kernel-based equiprobabilistic topographic map formation. Neural Computation, 10, pp [5] Van Hulle, M.M. (1998). Clustering with kernel-based equiprobabilistic topographic maps. Proc. IEEE NNSP98 (Cambridge, U.K.), pp [6] Watanabe, H., Matsumoto, Y., Tanaka, S., & Katagiri, S. (1998). Sound Monitoring based on the generalized probabilistic descent method. Proc. IEEE NNSP98 (Cambridge, U.K.), pp [7] Xie, F., and Van Compernolle, D. (1994). A family of MLP based nonlinear spectral estimators for noise reduction. Proc. ICASSP'94, vol. II, pp
Independent Component Analysis and Unsupervised Learning. Jen-Tzung Chien
Independent Component Analysis and Unsupervised Learning Jen-Tzung Chien TABLE OF CONTENTS 1. Independent Component Analysis 2. Case Study I: Speech Recognition Independent voices Nonparametric likelihood
More informationIndependent Component Analysis and Unsupervised Learning
Independent Component Analysis and Unsupervised Learning Jen-Tzung Chien National Cheng Kung University TABLE OF CONTENTS 1. Independent Component Analysis 2. Case Study I: Speech Recognition Independent
More informationSINGLE CHANNEL SPEECH MUSIC SEPARATION USING NONNEGATIVE MATRIX FACTORIZATION AND SPECTRAL MASKS. Emad M. Grais and Hakan Erdogan
SINGLE CHANNEL SPEECH MUSIC SEPARATION USING NONNEGATIVE MATRIX FACTORIZATION AND SPECTRAL MASKS Emad M. Grais and Hakan Erdogan Faculty of Engineering and Natural Sciences, Sabanci University, Orhanli
More informationElec4621 Advanced Digital Signal Processing Chapter 11: Time-Frequency Analysis
Elec461 Advanced Digital Signal Processing Chapter 11: Time-Frequency Analysis Dr. D. S. Taubman May 3, 011 In this last chapter of your notes, we are interested in the problem of nding the instantaneous
More informationMULTIVARIATE EDGEWORTH-BASED ENTROPY ESTIMATION. Marc M. Van Hulle
MULTIARIATE EDGEWORTH-BASED ENTROPY ESTIMATION Marc M. an Hulle K.U.Leuven, Laboratorium voor Neuro- en Psychofysiologie Campus Gasthuisberg, Herestraat 9, B-000 Leuven, BELGIUM E-mail: marc@neuro.kuleuven.ac.be
More informationNMF WITH SPECTRAL AND TEMPORAL CONTINUITY CRITERIA FOR MONAURAL SOUND SOURCE SEPARATION. Julian M. Becker, Christian Sohn and Christian Rohlfing
NMF WITH SPECTRAL AND TEMPORAL CONTINUITY CRITERIA FOR MONAURAL SOUND SOURCE SEPARATION Julian M. ecker, Christian Sohn Christian Rohlfing Institut für Nachrichtentechnik RWTH Aachen University D-52056
More informationSignal Modeling Techniques in Speech Recognition. Hassan A. Kingravi
Signal Modeling Techniques in Speech Recognition Hassan A. Kingravi Outline Introduction Spectral Shaping Spectral Analysis Parameter Transforms Statistical Modeling Discussion Conclusions 1: Introduction
More informationMULTI-RESOLUTION SIGNAL DECOMPOSITION WITH TIME-DOMAIN SPECTROGRAM FACTORIZATION. Hirokazu Kameoka
MULTI-RESOLUTION SIGNAL DECOMPOSITION WITH TIME-DOMAIN SPECTROGRAM FACTORIZATION Hiroazu Kameoa The University of Toyo / Nippon Telegraph and Telephone Corporation ABSTRACT This paper proposes a novel
More informationRemaining energy on log scale Number of linear PCA components
NONLINEAR INDEPENDENT COMPONENT ANALYSIS USING ENSEMBLE LEARNING: EXPERIMENTS AND DISCUSSION Harri Lappalainen, Xavier Giannakopoulos, Antti Honkela, and Juha Karhunen Helsinki University of Technology,
More informationDETECTING PROCESS STATE CHANGES BY NONLINEAR BLIND SOURCE SEPARATION. Alexandre Iline, Harri Valpola and Erkki Oja
DETECTING PROCESS STATE CHANGES BY NONLINEAR BLIND SOURCE SEPARATION Alexandre Iline, Harri Valpola and Erkki Oja Laboratory of Computer and Information Science Helsinki University of Technology P.O.Box
More informationUsing Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method
Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method Antti Honkela 1, Stefan Harmeling 2, Leo Lundqvist 1, and Harri Valpola 1 1 Helsinki University of Technology,
More informationNon-negative Matrix Factor Deconvolution; Extraction of Multiple Sound Sources from Monophonic Inputs
MITSUBISHI ELECTRIC RESEARCH LABORATORIES http://www.merl.com Non-negative Matrix Factor Deconvolution; Extraction of Multiple Sound Sources from Monophonic Inputs Paris Smaragdis TR2004-104 September
More informationy(n) Time Series Data
Recurrent SOM with Local Linear Models in Time Series Prediction Timo Koskela, Markus Varsta, Jukka Heikkonen, and Kimmo Kaski Helsinki University of Technology Laboratory of Computational Engineering
More information2D Spectrogram Filter for Single Channel Speech Enhancement
Proceedings of the 7th WSEAS International Conference on Signal, Speech and Image Processing, Beijing, China, September 15-17, 007 89 D Spectrogram Filter for Single Channel Speech Enhancement HUIJUN DING,
More informationEECS490: Digital Image Processing. Lecture #26
Lecture #26 Moments; invariant moments Eigenvector, principal component analysis Boundary coding Image primitives Image representation: trees, graphs Object recognition and classes Minimum distance classifiers
More informationSingle Channel Signal Separation Using MAP-based Subspace Decomposition
Single Channel Signal Separation Using MAP-based Subspace Decomposition Gil-Jin Jang, Te-Won Lee, and Yung-Hwan Oh 1 Spoken Language Laboratory, Department of Computer Science, KAIST 373-1 Gusong-dong,
More informationSingle Channel Music Sound Separation Based on Spectrogram Decomposition and Note Classification
Single Channel Music Sound Separation Based on Spectrogram Decomposition and Note Classification Hafiz Mustafa and Wenwu Wang Centre for Vision, Speech and Signal Processing (CVSSP) University of Surrey,
More informationNon-Negative Matrix Factorization And Its Application to Audio. Tuomas Virtanen Tampere University of Technology
Non-Negative Matrix Factorization And Its Application to Audio Tuomas Virtanen Tampere University of Technology tuomas.virtanen@tut.fi 2 Contents Introduction to audio signals Spectrogram representation
More informationORIENTED PCA AND BLIND SIGNAL SEPARATION
ORIENTED PCA AND BLIND SIGNAL SEPARATION K. I. Diamantaras Department of Informatics TEI of Thessaloniki Sindos 54101, Greece kdiamant@it.teithe.gr Th. Papadimitriou Department of Int. Economic Relat.
More informationSYSTEM RECONSTRUCTION FROM SELECTED HOS REGIONS. Haralambos Pozidis and Athina P. Petropulu. Drexel University, Philadelphia, PA 19104
SYSTEM RECOSTRUCTIO FROM SELECTED HOS REGIOS Haralambos Pozidis and Athina P. Petropulu Electrical and Computer Engineering Department Drexel University, Philadelphia, PA 94 Tel. (25) 895-2358 Fax. (25)
More informationA Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier
A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier Seiichi Ozawa, Shaoning Pang, and Nikola Kasabov Graduate School of Science and Technology, Kobe
More informationTHE PROBLEMS OF ROBUST LPC PARAMETRIZATION FOR. Petr Pollak & Pavel Sovka. Czech Technical University of Prague
THE PROBLEMS OF ROBUST LPC PARAMETRIZATION FOR SPEECH CODING Petr Polla & Pavel Sova Czech Technical University of Prague CVUT FEL K, 66 7 Praha 6, Czech Republic E-mail: polla@noel.feld.cvut.cz Abstract
More information1 Introduction Blind source separation (BSS) is a fundamental problem which is encountered in a variety of signal processing problems where multiple s
Blind Separation of Nonstationary Sources in Noisy Mixtures Seungjin CHOI x1 and Andrzej CICHOCKI y x Department of Electrical Engineering Chungbuk National University 48 Kaeshin-dong, Cheongju Chungbuk
More informationREAL-TIME TIME-FREQUENCY BASED BLIND SOURCE SEPARATION. Scott Rickard, Radu Balan, Justinian Rosca. Siemens Corporate Research Princeton, NJ 08540
REAL-TIME TIME-FREQUENCY BASED BLIND SOURCE SEPARATION Scott Rickard, Radu Balan, Justinian Rosca Siemens Corporate Research Princeton, NJ 84 fscott.rickard,radu.balan,justinian.roscag@scr.siemens.com
More informationSelf-Organization by Optimizing Free-Energy
Self-Organization by Optimizing Free-Energy J.J. Verbeek, N. Vlassis, B.J.A. Kröse University of Amsterdam, Informatics Institute Kruislaan 403, 1098 SJ Amsterdam, The Netherlands Abstract. We present
More informationNOISE ROBUST RELATIVE TRANSFER FUNCTION ESTIMATION. M. Schwab, P. Noll, and T. Sikora. Technical University Berlin, Germany Communication System Group
NOISE ROBUST RELATIVE TRANSFER FUNCTION ESTIMATION M. Schwab, P. Noll, and T. Sikora Technical University Berlin, Germany Communication System Group Einsteinufer 17, 1557 Berlin (Germany) {schwab noll
More informationSPEECH ANALYSIS AND SYNTHESIS
16 Chapter 2 SPEECH ANALYSIS AND SYNTHESIS 2.1 INTRODUCTION: Speech signal analysis is used to characterize the spectral information of an input speech signal. Speech signal analysis [52-53] techniques
More informationNon-Euclidean Independent Component Analysis and Oja's Learning
Non-Euclidean Independent Component Analysis and Oja's Learning M. Lange 1, M. Biehl 2, and T. Villmann 1 1- University of Appl. Sciences Mittweida - Dept. of Mathematics Mittweida, Saxonia - Germany 2-
More informationARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92
ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92 BIOLOGICAL INSPIRATIONS Some numbers The human brain contains about 10 billion nerve cells (neurons) Each neuron is connected to the others through 10000
More informationUnsupervised learning: beyond simple clustering and PCA
Unsupervised learning: beyond simple clustering and PCA Liza Rebrova Self organizing maps (SOM) Goal: approximate data points in R p by a low-dimensional manifold Unlike PCA, the manifold does not have
More informationComparative Performance Analysis of Three Algorithms for Principal Component Analysis
84 R. LANDQVIST, A. MOHAMMED, COMPARATIVE PERFORMANCE ANALYSIS OF THR ALGORITHMS Comparative Performance Analysis of Three Algorithms for Principal Component Analysis Ronnie LANDQVIST, Abbas MOHAMMED Dept.
More informationBlind Spectral-GMM Estimation for Underdetermined Instantaneous Audio Source Separation
Blind Spectral-GMM Estimation for Underdetermined Instantaneous Audio Source Separation Simon Arberet 1, Alexey Ozerov 2, Rémi Gribonval 1, and Frédéric Bimbot 1 1 METISS Group, IRISA-INRIA Campus de Beaulieu,
More informationMITIGATING UNCORRELATED PERIODIC DISTURBANCE IN NARROWBAND ACTIVE NOISE CONTROL SYSTEMS
17th European Signal Processing Conference (EUSIPCO 29) Glasgow, Scotland, August 24-28, 29 MITIGATING UNCORRELATED PERIODIC DISTURBANCE IN NARROWBAND ACTIVE NOISE CONTROL SYSTEMS Muhammad Tahir AKHTAR
More informationPower Supply Quality Analysis Using S-Transform and SVM Classifier
Journal of Power and Energy Engineering, 2014, 2, 438-447 Published Online April 2014 in SciRes. http://www.scirp.org/journal/jpee http://dx.doi.org/10.4236/jpee.2014.24059 Power Supply Quality Analysis
More informationwhere A 2 IR m n is the mixing matrix, s(t) is the n-dimensional source vector (n» m), and v(t) is additive white noise that is statistically independ
BLIND SEPARATION OF NONSTATIONARY AND TEMPORALLY CORRELATED SOURCES FROM NOISY MIXTURES Seungjin CHOI x and Andrzej CICHOCKI y x Department of Electrical Engineering Chungbuk National University, KOREA
More informationA Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement
A Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement Simon Leglaive 1 Laurent Girin 1,2 Radu Horaud 1 1: Inria Grenoble Rhône-Alpes 2: Univ. Grenoble Alpes, Grenoble INP,
More informationA Compressive Sensing Based Compressed Neural Network for Sound Source Localization
A Compressive Sensing Based Compressed Neural Network for Sound Source Localization Mehdi Banitalebi Dehkordi Speech Processing Research Lab Yazd University Yazd, Iran mahdi_banitalebi@stu.yazduni.ac.ir
More informationCovariance smoothing and consistent Wiener filtering for artifact reduction in audio source separation
Covariance smoothing and consistent Wiener filtering for artifact reduction in audio source separation Emmanuel Vincent METISS Team Inria Rennes - Bretagne Atlantique E. Vincent (Inria) Artifact reduction
More informationA Robust PCA by LMSER Learning with Iterative Error. Bai-ling Zhang Irwin King Lei Xu.
A Robust PCA by LMSER Learning with Iterative Error Reinforcement y Bai-ling Zhang Irwin King Lei Xu blzhang@cs.cuhk.hk king@cs.cuhk.hk lxu@cs.cuhk.hk Department of Computer Science The Chinese University
More informationOptimal Speech Enhancement Under Signal Presence Uncertainty Using Log-Spectral Amplitude Estimator
1 Optimal Speech Enhancement Under Signal Presence Uncertainty Using Log-Spectral Amplitude Estimator Israel Cohen Lamar Signal Processing Ltd. P.O.Box 573, Yokneam Ilit 20692, Israel E-mail: icohen@lamar.co.il
More informationLearning Vector Quantization
Learning Vector Quantization Neural Computation : Lecture 18 John A. Bullinaria, 2015 1. SOM Architecture and Algorithm 2. Vector Quantization 3. The Encoder-Decoder Model 4. Generalized Lloyd Algorithms
More informationFiltered-X LMS vs repetitive control for active structural acoustic control of periodic disturbances
Filtered-X LMS vs repetitive control for active structural acoustic control of periodic disturbances B. Stallaert 1, G. Pinte 2, S. Devos 2, W. Symens 2, J. Swevers 1, P. Sas 1 1 K.U.Leuven, Department
More informationGaussian Processes for Audio Feature Extraction
Gaussian Processes for Audio Feature Extraction Dr. Richard E. Turner (ret26@cam.ac.uk) Computational and Biological Learning Lab Department of Engineering University of Cambridge Machine hearing pipeline
More informationReview: Learning Bimodal Structures in Audio-Visual Data
Review: Learning Bimodal Structures in Audio-Visual Data CSE 704 : Readings in Joint Visual, Lingual and Physical Models and Inference Algorithms Suren Kumar Vision and Perceptual Machines Lab 106 Davis
More informationMusical noise reduction in time-frequency-binary-masking-based blind source separation systems
Musical noise reduction in time-frequency-binary-masing-based blind source separation systems, 3, a J. Čermá, 1 S. Arai, 1. Sawada and 1 S. Maino 1 Communication Science Laboratories, Corporation, Kyoto,
More informationx 1 (t) Spectrogram t s
A METHOD OF ICA IN TIME-FREQUENCY DOMAIN Shiro Ikeda PRESTO, JST Hirosawa 2-, Wako, 35-98, Japan Shiro.Ikeda@brain.riken.go.jp Noboru Murata RIKEN BSI Hirosawa 2-, Wako, 35-98, Japan Noboru.Murata@brain.riken.go.jp
More information1 Introduction Independent component analysis (ICA) [10] is a statistical technique whose main applications are blind source separation, blind deconvo
The Fixed-Point Algorithm and Maximum Likelihood Estimation for Independent Component Analysis Aapo Hyvarinen Helsinki University of Technology Laboratory of Computer and Information Science P.O.Box 5400,
More informationESTIMATION OF RELATIVE TRANSFER FUNCTION IN THE PRESENCE OF STATIONARY NOISE BASED ON SEGMENTAL POWER SPECTRAL DENSITY MATRIX SUBTRACTION
ESTIMATION OF RELATIVE TRANSFER FUNCTION IN THE PRESENCE OF STATIONARY NOISE BASED ON SEGMENTAL POWER SPECTRAL DENSITY MATRIX SUBTRACTION Xiaofei Li 1, Laurent Girin 1,, Radu Horaud 1 1 INRIA Grenoble
More informationProc. of NCC 2010, Chennai, India
Proc. of NCC 2010, Chennai, India Trajectory and surface modeling of LSF for low rate speech coding M. Deepak and Preeti Rao Department of Electrical Engineering Indian Institute of Technology, Bombay
More informationDesign Criteria for the Quadratically Interpolated FFT Method (I): Bias due to Interpolation
CENTER FOR COMPUTER RESEARCH IN MUSIC AND ACOUSTICS DEPARTMENT OF MUSIC, STANFORD UNIVERSITY REPORT NO. STAN-M-4 Design Criteria for the Quadratically Interpolated FFT Method (I): Bias due to Interpolation
More information"Robust Automatic Speech Recognition through on-line Semi Blind Source Extraction"
"Robust Automatic Speech Recognition through on-line Semi Blind Source Extraction" Francesco Nesta, Marco Matassoni {nesta, matassoni}@fbk.eu Fondazione Bruno Kessler-Irst, Trento (ITALY) For contacts:
More informationAPPLICATION OF MVDR BEAMFORMING TO SPHERICAL ARRAYS
AMBISONICS SYMPOSIUM 29 June 2-27, Graz APPLICATION OF MVDR BEAMFORMING TO SPHERICAL ARRAYS Anton Schlesinger 1, Marinus M. Boone 2 1 University of Technology Delft, The Netherlands (a.schlesinger@tudelft.nl)
More informationAN INVERTIBLE DISCRETE AUDITORY TRANSFORM
COMM. MATH. SCI. Vol. 3, No. 1, pp. 47 56 c 25 International Press AN INVERTIBLE DISCRETE AUDITORY TRANSFORM JACK XIN AND YINGYONG QI Abstract. A discrete auditory transform (DAT) from sound signal to
More informationEquivalence Probability and Sparsity of Two Sparse Solutions in Sparse Representation
IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 19, NO. 12, DECEMBER 2008 2009 Equivalence Probability and Sparsity of Two Sparse Solutions in Sparse Representation Yuanqing Li, Member, IEEE, Andrzej Cichocki,
More informationMULTISENSORY SPEECH ENHANCEMENT IN NOISY ENVIRONMENTS USING BONE-CONDUCTED AND AIR-CONDUCTED MICROPHONES. Mingzi Li,Israel Cohen and Saman Mousazadeh
MULTISENSORY SPEECH ENHANCEMENT IN NOISY ENVIRONMENTS USING BONE-CONDUCTED AND AIR-CONDUCTED MICROPHONES Mingzi Li,Israel Cohen and Saman Mousazadeh Department of Electrical Engineering, Technion - Israel
More informationCOMPARISON OF TWO FEATURE EXTRACTION METHODS BASED ON MAXIMIZATION OF MUTUAL INFORMATION. Nikolay Chumerin, Marc M. Van Hulle
COMPARISON OF TWO FEATURE EXTRACTION METHODS BASED ON MAXIMIZATION OF MUTUAL INFORMATION Nikolay Chumerin, Marc M. Van Hulle K.U.Leuven, Laboratorium voor Neuro- en Psychofysiologie Campus Gasthuisberg,
More informationA new fast algorithm for blind MA-system identication. based on higher order cumulants. K.D. Kammeyer and B. Jelonnek
SPIE Advanced Signal Proc: Algorithms, Architectures & Implementations V, San Diego, -9 July 99 A new fast algorithm for blind MA-system identication based on higher order cumulants KD Kammeyer and B Jelonnek
More informationBayesian ensemble learning of generative models
Chapter Bayesian ensemble learning of generative models Harri Valpola, Antti Honkela, Juha Karhunen, Tapani Raiko, Xavier Giannakopoulos, Alexander Ilin, Erkki Oja 65 66 Bayesian ensemble learning of generative
More informationProbabilistic Inference of Speech Signals from Phaseless Spectrograms
Probabilistic Inference of Speech Signals from Phaseless Spectrograms Kannan Achan, Sam T. Roweis, Brendan J. Frey Machine Learning Group University of Toronto Abstract Many techniques for complex speech
More informationSubmitted to Electronics Letters. Indexing terms: Signal Processing, Adaptive Filters. The Combined LMS/F Algorithm Shao-Jen Lim and John G. Harris Co
Submitted to Electronics Letters. Indexing terms: Signal Processing, Adaptive Filters. The Combined LMS/F Algorithm Shao-Jen Lim and John G. Harris Computational Neuro-Engineering Laboratory University
More informationw 1 output input &N 1 x w n w N =&2
ISSN 98-282 Technical Report L Noise Suppression in Training Data for Improving Generalization Akiko Nakashima, Akira Hirabayashi, and Hidemitsu OGAWA TR96-9 November Department of Computer Science Tokyo
More informationNotes on Latent Semantic Analysis
Notes on Latent Semantic Analysis Costas Boulis 1 Introduction One of the most fundamental problems of information retrieval (IR) is to find all documents (and nothing but those) that are semantically
More informationA comparative study of time-delay estimation techniques for convolutive speech mixtures
A comparative study of time-delay estimation techniques for convolutive speech mixtures COSME LLERENA AGUILAR University of Alcala Signal Theory and Communications 28805 Alcalá de Henares SPAIN cosme.llerena@uah.es
More informationNew Statistical Model for the Enhancement of Noisy Speech
New Statistical Model for the Enhancement of Noisy Speech Electrical Engineering Department Technion - Israel Institute of Technology February 22, 27 Outline Problem Formulation and Motivation 1 Problem
More informationIngo Ahrns, Jorg Bruske, Gerald Sommer. Christian Albrechts University of Kiel - Cognitive Systems Group. Preusserstr Kiel - Germany
On-line Learning with Dynamic Cell Structures Ingo Ahrns, Jorg Bruske, Gerald Sommer Christian Albrechts University of Kiel - Cognitive Systems Group Preusserstr. 1-9 - 24118 Kiel - Germany Phone: ++49
More informationBlind Channel Equalization in Impulse Noise
Blind Channel Equalization in Impulse Noise Rubaiyat Yasmin and Tetsuya Shimamura Graduate School of Science and Engineering, Saitama University 255 Shimo-okubo, Sakura-ku, Saitama 338-8570, Japan yasmin@sie.ics.saitama-u.ac.jp
More informationHST.582J / 6.555J / J Biomedical Signal and Image Processing Spring 2007
MIT OpenCourseWare http://ocw.mit.edu HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing Spring 2007 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
More informationNeural Networks for Two-Group Classification Problems with Monotonicity Hints
Neural Networks for Two-Group Classification Problems with Monotonicity Hints P. Lory 1, D. Gietl Institut für Wirtschaftsinformatik, Universität Regensburg, D-93040 Regensburg, Germany Abstract: Neural
More informationA Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier
A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier Seiichi Ozawa 1, Shaoning Pang 2, and Nikola Kasabov 2 1 Graduate School of Science and Technology,
More informationAnalysis of audio intercepts: Can we identify and locate the speaker?
Motivation Analysis of audio intercepts: Can we identify and locate the speaker? K V Vijay Girish, PhD Student Research Advisor: Prof A G Ramakrishnan Research Collaborator: Dr T V Ananthapadmanabha Medical
More informationChirp Transform for FFT
Chirp Transform for FFT Since the FFT is an implementation of the DFT, it provides a frequency resolution of 2π/N, where N is the length of the input sequence. If this resolution is not sufficient in a
More informationA METHOD OF ADAPTATION BETWEEN STEEPEST- DESCENT AND NEWTON S ALGORITHM FOR MULTI- CHANNEL ACTIVE CONTROL OF TONAL NOISE AND VIBRATION
A METHOD OF ADAPTATION BETWEEN STEEPEST- DESCENT AND NEWTON S ALGORITHM FOR MULTI- CHANNEL ACTIVE CONTROL OF TONAL NOISE AND VIBRATION Jordan Cheer and Stephen Daley Institute of Sound and Vibration Research,
More informationProceedings of Meetings on Acoustics
Proceedings of Meetings on Acoustics Volume 19, 2013 http://acousticalsociety.org/ ICA 2013 Montreal Montreal, Canada 2-7 June 2013 Architectural Acoustics Session 1pAAa: Advanced Analysis of Room Acoustics:
More informationAnalysis of Interest Rate Curves Clustering Using Self-Organising Maps
Analysis of Interest Rate Curves Clustering Using Self-Organising Maps M. Kanevski (1), V. Timonin (1), A. Pozdnoukhov(1), M. Maignan (1,2) (1) Institute of Geomatics and Analysis of Risk (IGAR), University
More informationAuto-correlation of retinal ganglion cell mosaics shows hexagonal structure
Supplementary Discussion Auto-correlation of retinal ganglion cell mosaics shows hexagonal structure Wässle and colleagues first observed that the local structure of cell mosaics was approximately hexagonal
More informationNeural Networks and Machine Learning research at the Laboratory of Computer and Information Science, Helsinki University of Technology
Neural Networks and Machine Learning research at the Laboratory of Computer and Information Science, Helsinki University of Technology Erkki Oja Department of Computer Science Aalto University, Finland
More informationMachine Learning for Signal Processing Sparse and Overcomplete Representations
Machine Learning for Signal Processing Sparse and Overcomplete Representations Abelino Jimenez (slides from Bhiksha Raj and Sourish Chaudhuri) Oct 1, 217 1 So far Weights Data Basis Data Independent ICA
More informationError Empirical error. Generalization error. Time (number of iteration)
Submitted to Neural Networks. Dynamics of Batch Learning in Multilayer Networks { Overrealizability and Overtraining { Kenji Fukumizu The Institute of Physical and Chemical Research (RIKEN) E-mail: fuku@brain.riken.go.jp
More informationIn: Proc. BENELEARN-98, 8th Belgian-Dutch Conference on Machine Learning, pp 9-46, 998 Linear Quadratic Regulation using Reinforcement Learning Stephan ten Hagen? and Ben Krose Department of Mathematics,
More informationConsistent Wiener Filtering for Audio Source Separation
MITSUBISHI ELECTRIC RESEARCH LABORATORIES http://www.merl.com Consistent Wiener Filtering for Audio Source Separation Le Roux, J.; Vincent, E. TR2012-090 October 2012 Abstract Wiener filtering is one of
More informationOn Spectral Basis Selection for Single Channel Polyphonic Music Separation
On Spectral Basis Selection for Single Channel Polyphonic Music Separation Minje Kim and Seungjin Choi Department of Computer Science Pohang University of Science and Technology San 31 Hyoja-dong, Nam-gu
More informationOVERLAPPING ANIMAL SOUND CLASSIFICATION USING SPARSE REPRESENTATION
OVERLAPPING ANIMAL SOUND CLASSIFICATION USING SPARSE REPRESENTATION Na Lin, Haixin Sun Xiamen University Key Laboratory of Underwater Acoustic Communication and Marine Information Technology, Ministry
More informationSparse filter models for solving permutation indeterminacy in convolutive blind source separation
Sparse filter models for solving permutation indeterminacy in convolutive blind source separation Prasad Sudhakar, Rémi Gribonval To cite this version: Prasad Sudhakar, Rémi Gribonval. Sparse filter models
More informationFUNDAMENTAL LIMITATION OF FREQUENCY DOMAIN BLIND SOURCE SEPARATION FOR CONVOLVED MIXTURE OF SPEECH
FUNDAMENTAL LIMITATION OF FREQUENCY DOMAIN BLIND SOURCE SEPARATION FOR CONVOLVED MIXTURE OF SPEECH Shoko Araki Shoji Makino Ryo Mukai Tsuyoki Nishikawa Hiroshi Saruwatari NTT Communication Science Laboratories
More informationUndercomplete Independent Component. Analysis for Signal Separation and. Dimension Reduction. Category: Algorithms and Architectures.
Undercomplete Independent Component Analysis for Signal Separation and Dimension Reduction John Porrill and James V Stone Psychology Department, Sheeld University, Sheeld, S10 2UR, England. Tel: 0114 222
More informationTWO METHODS FOR ESTIMATING OVERCOMPLETE INDEPENDENT COMPONENT BASES. Mika Inki and Aapo Hyvärinen
TWO METHODS FOR ESTIMATING OVERCOMPLETE INDEPENDENT COMPONENT BASES Mika Inki and Aapo Hyvärinen Neural Networks Research Centre Helsinki University of Technology P.O. Box 54, FIN-215 HUT, Finland ABSTRACT
More informationSHIFT-INVARIANT DICTIONARY LEARNING FOR SPARSE REPRESENTATIONS: EXTENDING K-SVD
SHIFT-INVARIANT DICTIONARY LEARNING FOR SPARSE REPRESENTATIONS: EXTENDING K-SVD Boris Mailhé, Sylvain Lesage,Rémi Gribonval and Frédéric Bimbot Projet METISS Centre de Recherche INRIA Rennes - Bretagne
More informationChapter 9. Linear Predictive Analysis of Speech Signals 语音信号的线性预测分析
Chapter 9 Linear Predictive Analysis of Speech Signals 语音信号的线性预测分析 1 LPC Methods LPC methods are the most widely used in speech coding, speech synthesis, speech recognition, speaker recognition and verification
More informationConvolutive Blind Source Separation based on Multiple Decorrelation. Lucas Parra, Clay Spence, Bert De Vries Sarno Corporation, CN-5300, Princeton, NJ
Convolutive Blind Source Separation based on Multiple Decorrelation. Lucas Parra, Clay Spence, Bert De Vries Sarno Corporation, CN-5300, Princeton, NJ 08543 lparra j cspence j bdevries @ sarno.com Abstract
More informationHST.582J/6.555J/16.456J
Blind Source Separation: PCA & ICA HST.582J/6.555J/16.456J Gari D. Clifford gari [at] mit. edu http://www.mit.edu/~gari G. D. Clifford 2005-2009 What is BSS? Assume an observation (signal) is a linear
More informationTone Analysis in Harmonic-Frequency Domain and Feature Reduction using KLT+LVQ for Thai Isolated Word Recognition
Tone Analysis in Harmonic-Frequency Domain and Feature Reduction using KLT+LVQ for Thai Isolated Word Recognition SARITCHAI PREDAWAN 1 PRASIT JIYAPANICHKUL 2 and CHOM KIMPAN 3 Faculty of Information Technology
More informationBobby Hunt, Mariappan S. Nadar, Paul Keller, Eric VonColln, and Anupam Goyal III. ASSOCIATIVE RECALL BY A POLYNOMIAL MAPPING
Synthesis of a Nonrecurrent Associative Memory Model Based on a Nonlinear Transformation in the Spectral Domain p. 1 Bobby Hunt, Mariappan S. Nadar, Paul Keller, Eric VonColln, Anupam Goyal Abstract -
More informationx 104
Departement Elektrotechniek ESAT-SISTA/TR 98-3 Identication of the circulant modulated Poisson process: a time domain approach Katrien De Cock, Tony Van Gestel and Bart De Moor 2 April 998 Submitted for
More informationReading Group on Deep Learning Session 1
Reading Group on Deep Learning Session 1 Stephane Lathuiliere & Pablo Mesejo 2 June 2016 1/31 Contents Introduction to Artificial Neural Networks to understand, and to be able to efficiently use, the popular
More informationNONLINEAR INDEPENDENT FACTOR ANALYSIS BY HIERARCHICAL MODELS
NONLINEAR INDEPENDENT FACTOR ANALYSIS BY HIERARCHICAL MODELS Harri Valpola, Tomas Östman and Juha Karhunen Helsinki University of Technology, Neural Networks Research Centre P.O. Box 5400, FIN-02015 HUT,
More informationA Subspace Approach to Estimation of. Measurements 1. Carlos E. Davila. Electrical Engineering Department, Southern Methodist University
EDICS category SP 1 A Subspace Approach to Estimation of Autoregressive Parameters From Noisy Measurements 1 Carlos E Davila Electrical Engineering Department, Southern Methodist University Dallas, Texas
More informationAlpha-Stable Distributions in Signal Processing of Audio Signals
Alpha-Stable Distributions in Signal Processing of Audio Signals Preben Kidmose, Department of Mathematical Modelling, Section for Digital Signal Processing, Technical University of Denmark, Building 3,
More informationSTRUCTURE-AWARE DICTIONARY LEARNING WITH HARMONIC ATOMS
19th European Signal Processing Conference (EUSIPCO 2011) Barcelona, Spain, August 29 - September 2, 2011 STRUCTURE-AWARE DICTIONARY LEARNING WITH HARMONIC ATOMS Ken O Hanlon and Mark D.Plumbley Queen
More informationPROPERTIES OF THE EMPIRICAL CHARACTERISTIC FUNCTION AND ITS APPLICATION TO TESTING FOR INDEPENDENCE. Noboru Murata
' / PROPERTIES OF THE EMPIRICAL CHARACTERISTIC FUNCTION AND ITS APPLICATION TO TESTING FOR INDEPENDENCE Noboru Murata Waseda University Department of Electrical Electronics and Computer Engineering 3--
More informationSound Source Tracking Using Microphone Arrays
Sound Source Tracking Using Microphone Arrays WANG PENG and WEE SER Center for Signal Processing School of Electrical & Electronic Engineering Nanayang Technological Univerisy SINGAPORE, 639798 Abstract:
More information