Apprentissage de représentations musicales à l'aide d'architectures profondes et multiéchelles

L'apprentissage machine (AM) est un outil important dans le domaine de la recherche d'information musicale (Music Information Retrieval ou MIR). De nombreuses tâches de MIR peuvent être résolues en entraînant un classifieur sur un ensemble de caractéristiques. Pour les tâches de MIR se basant sur l'audio musical, il est possible d'extraire de l'audio les caractéristiques pertinentes à l'aide de méthodes traitement de signal. Toutefois, certains aspects musicaux sont difficiles à extraire à l'aide de simples heuristiques. Afin d'obtenir des caractéristiques plus riches, il est possible d'utiliser l'AM pour apprendre une représentation musicale à partir de l'audio. Ces caractéristiques apprises permettent souvent d'améliorer la performance sur une tâche de MIR donnée. Afin d'apprendre des représentations musicales intéressantes, il est important de considérer les aspects particuliers à l'audio musical dans la conception des modèles d'apprentissage. Vu la structure temporelle et spectrale de l'audio musical, les représentations profondes et multiéchelles sont particulièrement bien conçues pour représenter la musique. Cette thèse porte sur l'apprentissage de représentations de l'audio musical. Des modèles profonds et multiéchelles améliorant l'état de l'art pour des tâches telles que la reconnaissance d'instrument, la reconnaissance de genre et l'étiquetage automatique y sont présentés.

Design and Development of Fiber Optic Sensors for Trace Detection of Certain Environmental Pollutants

In recent years,photonics has emerged as an essential technology related to such diverse fields like laser technology,fiber optics,communication,optical signal processing,computing,entertainment,consumer electronics etc.Availabilities of semiconductor lasers and low loss fibers have also revolutionized the field of sensor technology including telemetry. There exist fiber optic sensors which are sensitive,reliable.light weight and accurate devices which find applications in wide range of areas like biomedicine,aviation,surgery,pollution monitoring etc.,apart from areas in basic sciences.The present thesis deals with the design,fabrication and characterization of a variety of cost effective and sensitive fiber optic sensors for the trace detetction of certain environment pollutants in air and water.The sensor design is carried out using the techniques like evanescent waves,micro bending and long period gratings.

Photothermal beam deflection for non-destructive evaluation of semiconductor thin films

Non-destructive testing (NDT) is the use of non-invasive techniques to determine the integrity of a material, component, or structure. Engineers and scientists use NDT in a variety of applications, including medical imaging, materials analysis, and process control.Photothermal beam deflection technique is one of the most promising NDT technologies. Tremendous R&D effort has been made for improving the efficiency and simplicity of this technique. It is a popular technique because it can probe surfaces irrespective of the size of the sample and its surroundings. This technique has been used to characterize several semiconductor materials, because of its non-destructive and non-contact evaluation strategy. Its application further extends to analysis of wide variety of materials. Instrumentation of a NDT technique is very crucial for any material analysis. Chapter two explores the various excitation sources, source modulation techniques, detection and signal processing schemes currently practised. The features of the experimental arrangement including the steps for alignment, automation, data acquisition and data analysis are explained giving due importance to details.Theoretical studies form the backbone of photothermal techniques. The outcome of a theoretical work is the foundation of an application.The reliability of the theoretical model developed and used is proven from the studies done on crystalline.The technique is applied for analysis of transport properties such as thermal diffusivity, mobility, surface recombination velocity and minority carrier life time of the material and thermal imaging of solar cell absorber layer materials like CuInS2, CuInSe2 and SnS thin films.analysis of In2S3 thin films, which are used as buffer layer material in solar cells. The various influences of film composition, chlorine and silver incorporation in this material is brought out from the measurement of transport properties and analysis of sub band gap levels.The application of photothermal deflection technique for characterization of solar cells is a relatively new area that requires considerable attention.The application of photothermal deflection technique for characterization of solar cells is a relatively new area that requires considerable attention. Chapter six thus elucidates the theoretical aspects of application of photothermal techniques for solar cell analysis. The experimental design and method for determination of solar cell efficiency, optimum load resistance and series resistance with results from the analysis of CuInS2/In2S3 based solar cell forms the skeleton of this chapter.

Development of a New Transform:MRT

Fourier transform methods are employed heavily in digital signal processing. Discrete Fourier Transform (DFT) is among the most commonly used digital signal transforms. The exponential kernel of the DFT has the properties of symmetry and periodicity. Fast Fourier Transform (FFT) methods for fast DFT computation exploit these kernel properties in different ways. In this thesis, an approach of grouping data on the basis of the corresponding phase of the exponential kernel of the DFT is exploited to introduce a new digital signal transform, named the M-dimensional Real Transform (MRT), for l-D and 2-D signals. The new transform is developed using number theoretic principles as regards its specific features. A few properties of the transform are explored, and an inverse transform presented. A fundamental assumption is that the size of the input signal be even. The transform computation involves only real additions. The MRT is an integer-to-integer transform. There are two kinds of redundancy, complete redundancy & derived redundancy, in MRT. Redundancy is analyzed and removed to arrive at a more compact version called the Unique MRT (UMRT). l-D UMRT is a non-expansive transform for all signal sizes, while the 2-D UMRT is non-expansive for signal sizes that are powers of 2. The 2-D UMRT is applied in image processing applications like image compression and orientation analysis. The MRT & UMRT, being general transforms, will find potential applications in various fields of signal and image processing.

Modeling and Analysis of Some Heavy Tailed Time Series

The thesis has covered various aspects of modeling and analysis of finite mean time series with symmetric stable distributed innovations. Time series analysis based on Box and Jenkins methods are the most popular approaches where the models are linear and errors are Gaussian. We highlighted the limitations of classical time series analysis tools and explored some generalized tools and organized the approach parallel to the classical set up. In the present thesis we mainly studied the estimation and prediction of signal plus noise model. Here we assumed the signal and noise follow some models with symmetric stable innovations.We start the thesis with some motivating examples and application areas of alpha stable time series models. Classical time series analysis and corresponding theories based on finite variance models are extensively discussed in second chapter. We also surveyed the existing theories and methods correspond to infinite variance models in the same chapter. We present a linear filtering method for computing the filter weights assigned to the observation for estimating unobserved signal under general noisy environment in third chapter. Here we consider both the signal and the noise as stationary processes with infinite variance innovations. We derived semi infinite, double infinite and asymmetric signal extraction filters based on minimum dispersion criteria. Finite length filters based on Kalman-Levy filters are developed and identified the pattern of the filter weights. Simulation studies show that the proposed methods are competent enough in signal extraction for processes with infinite variance.Parameter estimation of autoregressive signals observed in a symmetric stable noise environment is discussed in fourth chapter. Here we used higher order Yule-Walker type estimation using auto-covariation function and exemplify the methods by simulation and application to Sea surface temperature data. We increased the number of Yule-Walker equations and proposed a ordinary least square estimate to the autoregressive parameters. Singularity problem of the auto-covariation matrix is addressed and derived a modified version of the Generalized Yule-Walker method using singular value decomposition.In fifth chapter of the thesis we introduced partial covariation function as a tool for stable time series analysis where covariance or partial covariance is ill defined. Asymptotic results of the partial auto-covariation is studied and its application in model identification of stable auto-regressive models are discussed. We generalize the Durbin-Levinson algorithm to include infinite variance models in terms of partial auto-covariation function and introduce a new information criteria for consistent order estimation of stable autoregressive model.In chapter six we explore the application of the techniques discussed in the previous chapter in signal processing. Frequency estimation of sinusoidal signal observed in symmetric stable noisy environment is discussed in this context. Here we introduced a parametric spectrum analysis and frequency estimate using power transfer function. Estimate of the power transfer function is obtained using the modified generalized Yule-Walker approach. Another important problem in statistical signal processing is to identify the number of sinusoidal components in an observed signal. We used a modified version of the proposed information criteria for this purpose.

MS Measurements

RMS measuring device is a nonlinear device consisting of linear and nonlinear devices. The performance of rms measurement is influenced by a number of factors; i) signal characteristics, 2) the measurement technique used and 3) the device characteristics. RMS measurement is not simple, particularly when the signals are complex and unknown. The problem of rms measurement on high crest-factor signals is fully discussed and a solution to this problem is presented in this thesis. The problem of rms measurement is systematically analized and found to have mainly three types of errors: (1) amplitude or waveform error 2) Frequency error and (3) averaging error. Various rms measurement techniques are studied and compared. On the basis of this study the rms -measurement is reclassified three categories: (1) Wave-form-error-free measurement (2) High-frequncy-error measurement and (3) Low-frequency error-free measurement. In modern digital sampled-data systems the signals are complex and waveform-error-free rms measurement is highly appreciated. Among the three basic blocks of rms measuring device the squarer is the most important one. A squaring technique is selected, that permits shaping of the squarer error characteristic in such a way as to achieve waveform-errob free rms measurement. The squarer is designed, fabricated and tested. A hybrid rms measurement using an analog rms computing device and digital display combines the speed of analog techniques and the resolution and ease of measurement of digital techniques. An A/D converter is modified to perform the square-rooting operation. A 10-V rms voltmeter using the developed rms detector is fabricated and tested. The chapters two, three and four analyse the problems involved in rms measurement and present a comparative study of rms computing techniques and devices. The fifth chapter gives the details of the developed rms detector that permits wave-form-error free rms measurement. The sixth chapter, enumerates the the highlights of the thesis and suggests a list of future projects

A novel Sigma–Delta based parallel analogue-to-residue converter

Animportant step in the residue number system(RNS) based signal processing is the conversion of signal into residue domain. Many implementations of this conversion have been proposed for various goals, and one of the implementations is by a direct conversion from an analogue input. A novel approach for analogue-to-residue conversion is proposed in this research using the most popular Sigma–Delta analogue-to-digital converter (SD-ADC). In this approach, the front end is the same as in traditional SD-ADC that uses Sigma–Delta (SD) modulator with appropriate dynamic range, but the filtering is doneby a filter implemented usingRNSarithmetic. Hence, the natural output of the filter is an RNS representation of the input signal. The resolution, conversion speed, hardware complexity and cost of implementation of the proposed SD based analogue-to-residue converter are compared with the existing analogue-to-residue converters based on Nyquist rate ADCs

DNA Sequence Representation methods

DNA sequence representation methods are used to denote a gene structure effectively and help in similarities/dissimilarities analysis of coding sequences. Many different kinds of representations have been proposed in the literature. They can be broadly classified into Numerical, Graphical, Geometrical and Hybrid representation methods. DNA structure and function analysis are made easy with graphical and geometrical representation methods since it gives visual representation of a DNA structure. In numerical method, numerical values are assigned to a sequence and digital signal processing methods are used to analyze the sequence. Hybrid approaches are also reported in the literature to analyze DNA sequences. This paper reviews the latest developments in DNA Sequence representation methods. We also present a taxonomy of various methods. A comparison of these methods where ever possible is also done

Wavelet based Independent Component Analysis for Multispectral Brain Tissue Classification

Multispectral analysis is a promising approach in tissue classification and abnormality detection from Magnetic Resonance (MR) images. But instability in accuracy and reproducibility of the classification results from conventional techniques keeps it far from clinical applications. Recent studies proposed Independent Component Analysis (ICA) as an effective method for source signals separation from multispectral MR data. However, it often fails to extract the local features like small abnormalities, especially from dependent real data. A multisignal wavelet analysis prior to ICA is proposed in this work to resolve these issues. Best de-correlated detail coefficients are combined with input images to give better classification results. Performance improvement of the proposed method over conventional ICA is effectively demonstrated by segmentation and classification using k-means clustering. Experimental results from synthetic and real data strongly confirm the positive effect of the new method with an improved Tanimoto index/Sensitivity values, 0.884/93.605, for reproduced small white matter lesions

Wideband Low-Distortion Sigma-Delta ADC for WLAN

This work presents a wideband low-distortion sigmadelta analog-to-digital converter (ADC) for Wireless Local Area Network (WLAN) standard. The proposed converter makes use of low-distortion swing suppression SDM architecture which is highly suitable for low oversampling ratios to attain high linearity over a wide bandwidth. The modulator employs a 2-2 cascaded sigma-delta modulator with feedforward path with a single-bit quantizer in the first stage and 4-bit in the second stage. The modulator is designed in TSMC 0.18um CMOS technology and operates at 1.8V supply voltage. Simulation results show that, a peak SNDR of 57dB and a spurious free dynamic range (SFDR) of 66dB is obtained for a 10MHz signal bandwidth, and an oversampling ratio of 8.

Developing a Child Friendly Text-to-Speech System

This paper discusses the implementation details of a child friendly, good quality, English text-to-speech (TTS) system that is phoneme-based, concatenative, easy to set up and use with little memory. Direct waveform concatenation and linear prediction coding (LPC) are used. Most existing TTS systems are unit-selection based, which use standard speech databases available in neutral adult voices.Here reduced memory is achieved by the concatenation of phonemes and by replacing phonetic wave files with their LPC coefficients. Linguistic analysis was used to reduce the algorithmic complexity instead of signal processing techniques. Sufficient degree of customization and generalization catering to the needs of the child user had been included through the provision for vocabulary and voice selection to suit the requisites of the child. Prosody had also been incorporated. This inexpensive TTS systemwas implemented inMATLAB, with the synthesis presented by means of a graphical user interface (GUI), thus making it child friendly. This can be used not only as an interesting language learning aid for the normal child but it also serves as a speech aid to the vocally disabled child. The quality of the synthesized speech was evaluated using the mean opinion score (MOS).

Discrete wavelet transform de-noising in eukaryotic gene splicing

This paper compares the most common digital signal processing methods of exon prediction in eukaryotes, and also proposes a technique for noise suppression in exon prediction. The specimen used here which has relevance in medical research, has been taken from the public genomic database - GenBank.Here exon prediction has been done using the digital signal processing methods viz. binary method, EIIP (electron-ion interaction psuedopotential) method and filter methods. Under filter method two filter designs, and two approaches using these two designs have been tried. The discrete wavelet transform has been used for de-noising of the exon plots.Results of exon prediction based on the methods mentioned above, which give values closest to the ones found in the NCBI database are given here. The exon plot de-noised using discrete wavelet transform is also given.Alterations to the proven methods as done by the authors, improves performance of exon prediction algorithms. Also it has been proven that the discrete wavelet transform is an effective tool for de-noising which can be used with exon prediction algorithms

Design, Development and Realization of Quadratic Volterra Filters for Selected Applications

The basic concepts of digital signal processing are taught to the students in engineering and science. The focus of the course is on linear, time invariant systems. The question as to what happens when the system is governed by a quadratic or cubic equation remains unanswered in the vast majority of literature on signal processing. Light has been shed on this problem when John V Mathews and Giovanni L Sicuranza published the book Polynomial Signal Processing. This book opened up an unseen vista of polynomial systems for signal and image processing. The book presented the theory and implementations of both adaptive and non-adaptive FIR and IIR quadratic systems which offer improved performance than conventional linear systems. The theory of quadratic systems presents a pristine and virgin area of research that offers computationally intensive work. Once the area of research is selected, the next issue is the choice of the software tool to carry out the work. Conventional languages like C and C++ are easily eliminated as they are not interpreted and lack good quality plotting libraries. MATLAB is proved to be very slow and so do SCILAB and Octave. The search for a language for scientific computing that was as fast as C, but with a good quality plotting library, ended up in Python, a distant relative of LISP. It proved to be ideal for scientific computing. An account of the use of Python, its scientific computing package scipy and the plotting library pylab is given in the appendix Initially, work is focused on designing predictors that exploit the polynomial nonlinearities inherent in speech generation mechanisms. Soon, the work got diverted into medical image processing which offered more potential to exploit by the use of quadratic methods. The major focus in this area is on quadratic edge detection methods for retinal images and fingerprints as well as de-noising raw MRI signals

Parameter Estimation in Chaotic Systems

This report examines how to estimate the parameters of a chaotic system given noisy observations of the state behavior of the system. Investigating parameter estimation for chaotic systems is interesting because of possible applications for high-precision measurement and for use in other signal processing, communication, and control applications involving chaotic systems. In this report, we examine theoretical issues regarding parameter estimation in chaotic systems and develop an efficient algorithm to perform parameter estimation. We discover two properties that are helpful for performing parameter estimation on non-structurally stable systems. First, it turns out that most data in a time series of state observations contribute very little information about the underlying parameters of a system, while a few sections of data may be extraordinarily sensitive to parameter changes. Second, for one-parameter families of systems, we demonstrate that there is often a preferred direction in parameter space governing how easily trajectories of one system can "shadow'" trajectories of nearby systems. This asymmetry of shadowing behavior in parameter space is proved for certain families of maps of the interval. Numerical evidence indicates that similar results may be true for a wide variety of other systems. Using the two properties cited above, we devise an algorithm for performing parameter estimation. Standard parameter estimation techniques such as the extended Kalman filter perform poorly on chaotic systems because of divergence problems. The proposed algorithm achieves accuracies several orders of magnitude better than the Kalman filter and has good convergence properties for large data sets.

Feature Point Detection and Curve Approximation for Early Processing of Freehand Sketches

Freehand sketching is both a natural and crucial part of design, yet is unsupported by current design automation software. We are working to combine the flexibility and ease of use of paper and pencil with the processing power of a computer to produce a design environment that feels as natural as paper, yet is considerably smarter. One of the most basic steps in accomplishing this is converting the original digitized pen strokes in the sketch into the intended geometric objects using feature point detection and approximation. We demonstrate how multiple sources of information can be combined for feature detection in strokes and apply this technique using two approaches to signal processing, one using simple average based thresholding and a second using scale space.