Dissertations / Theses on the topic 'Spectrogram'

In many areas of science, near-periodic phenomena represent important information within time-series data. This thesis takes the example of the detection of non-transitory frequency components in passive sonar data, a problem which finds many applications. This problem is typically transformed into the pattern recognition domain by representing the time-series data as a spectrogram, in which slowly varying periodic signals appear as curvilinear tracks. The research is initiated with a survey of the literature, which is focused upon research into the detection of tracks within spectrograms. An investigation into low-level feature detection reveals that none of the evaluated methods perform adequately within the low signal-to-noise ratios of real-life spectrograms and, therefore, two novel feature detectors are proposed. An investigation into the various sources of information available to the detection process shows that the most simple of these, the individual pixel intensity values, used by most existing algorithms, is not sufficient for the problem. To overcome these limitations, a novel low-level feature detector is integrated into a novel active contour track detection algorithm, and this serves to greatly increase detection rates at low signal-to-noise ratios. Furthermore, the algorithm integrates a priori knowledge of the harmonic process, which describes the relative positions of tracks, to augment the available information in difficult conditions. Empirical evaluation of the algorithm demonstrates that it is effective at detecting tracks at signal-to-noise ratios as low as: 0.5 dB with vertical; 3 dB with oblique; and 2 dB with sinusoidal variation of harmonic features. It is also concluded that the proposed potential energy increases the active contour's effectiveness in detecting all the track structures by a factor of eight (as determined by the line location accuracy measure), even at relatively high signal-to-noise ratios, and that incorporating a priori knowledge of the harmonic process increases the detection rate by a factor of two.

Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2016.
This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Cataloged from student-submitted PDF version of thesis.
Includes bibliographical references (pages 63-67).
I present the state-space multitaper approach for analyzing non-stationary time series. Nonstationary time series are commonly divided into small time windows for analysis, but existing methods lose predictive power by analyzing each window independently, even though nearby windows have similar spectral properties. The state-space multitaper algorithm combines two approaches for spectral analysis: the state-space approach models the relations between nearby windows, and the multitaper approach balances a bias-variance tradeoff inherent in Fourier analysis of finite interval data. I illustrate an application of the algorithm to real-time anesthesia monitoring, which could prevent traumatic cases of intraoperative awareness. I discuss issues including a real-time implementation and modeling the system's noise parameters. I identify the new problem of phase censorship, by which spectral leakage hides some information necessary to relate signal phases across windows in time.
by Michael K. Behr.
M. Eng.

Despite the recent successes of neural networks in a variety of domains, musical audio modeling is still considered a hard task, with features typically spanning tens of thousands of dimensions in input space. By formulating audio data compression as an unsupervised learning task, this project investigates the applicability of vector quantized neural network autoencoders for compressing spectrograms – image-like representations of audio. Using a recently proposed gradient-based method for approximating waveforms from reconstructed (real-valued) spectrograms, the discrete pipeline produces listenable reconstructions of surprising fidelity compared to uncompressed versions, even for out-of-domain examples. The results suggest that the learned discrete quantization method achieves about 9x harder spectrogram compression compared to its continuous counterpart, while achieving similar reconstructions, both qualitatively and in terms of quantitative error metrics.
Trots de senaste framgångarna för neurala nätverk på en rad olika områden är musikalisk ljudmodellering fortfarande en svår uppgift, med karakteristiska egenskaper som spänner över tiotusentals dimensioner i inputrymnden. Genom att formulera ljuddatakomprimering som en oövervakad inlärningsuppgift undersöker detta projekt användbarheten av vektorkvantiserade neurala nätverkbaserade självkodare på spektrogram – en bildliknande representation av ljud. Med en nyligen beskriven gradientbaserad metod för approximering av vågformer från rekonstruerade (realvärda) spektrogram, producerar den diskreta pipelinen lyssningsbara rekonstruktioner med överraskande ljudåtergivning jämfört med okomprimerade versioner, även för exempel utanför domänen. Resultaten tyder på att den lärda diskreta kvantiseringsmetoden uppnår ungefär nio gånger hårdare spektrogramkompression jämfört med sin kontinuerliga motsvarighet, samtidigt som den skapar liknande rekonstruktioner, både kvalitativt och enligt kvantitativa felmått.

Characterization of dispersion curves in plate-like structures is possible with guided Lamb waves. In this research, experimental development of dispersion curves relies on the spectrogram, which suffers from the Heisenberg Uncertainty Principle. Reassignment is capable of localizing ill--defined dispersion curves. Unfortunately, reassignment also introduces spurious components, which reduce reassignment performance. This research develops an algorithm that provides both localization of dispersion curves and elimination of spurious components. To achieve this, an alternative formulation of reassignment called differential reassignment is modified and superimposed with nonlinear anisotropic diffusion. This study first examines reassignment and diffusion components individually. Three different versions of differential reassignment are considered, two of which are modifications explicitly derived in this research. The combined algorithm is then applied to reassign experimentally measured spectrograms, leading to a significant increase in clarity and notch detection performance.

This thesis examines the application of the techniques of Fourier spectrogram and wavelet analysis to a low power embedded microprocessor application in a novel railway and rollingstock monitoring system. The safe and cost effective operation of freight railways is limited by the dynamic performance of wagons running on track. A monitoring system has been proposed comprising of low cost wireless sensing devices, dubbed “Health Cards”, to be installed on every wagon in the fleet. When marshalled into a train, the devices would sense accelerations and communicate via radio network to a master system in the locomotive. The integrated system would provide online information for decision support systems. Data throughput was heavily restricted by the network architecture, so significant signal analysis was required at the device level. An electronics engineering team at Central Queensland University developed a prototype Health Card, incorporating a 27MHz microcontroller and four dual axis accelerometers. A sensing arrangement and online analysis algorithms were required to detect and categorise dynamic events while operating within the constraints of the system. Time-frequency analysis reveals the time varying frequency content of signals, making it suitable to detect and characterise transient events. With efficient algorithms such as the Fast Fourier Transform, and Fast Wavelet Transform, time-frequency analysis methods can be implemented on a low power, embedded microcontroller. This thesis examines the application of time-frequency analysis techniques to wagon body acceleration signals, for the purpose of detecting poor dynamic performance of the wagon-track system. The Fourier spectrogram is implemented on the Health Card prototype and demonstrated in the laboratory. The research and algorithms provide a foundation for ongoing development as resources become available for system testing and validation.

We used a Byzantine Music piece performed by a well recognized chanter in order to derive experimentally the mean frequencies of the first five tones (D – A) of the diatonic scale of Byzantine Music. Then we compared the experimentally derived frequencies with frequencies proposed by two theoretical scales, both representative of traditional Byzantine Music chanting. We found that if a scale is performed by a traditional chanter is very close in frequency to the frequencies proposed theoretically. We then determined an allowed frequency deviation from the mean frequencies for each tone. The concept of allowed deviation is not provided by theory. Comparing our results to the notion of pitch discrimination from psychophysics we further established that the frequency differences are minute. The Attraction Effect was tested for a secondary tone (E) and the effect is quantified for the first time. The concept of the Attraction Effect is not explained in theory in terms of frequencies of tones.

O crescimento do volume de dados ocasionado pelo desenvolvimento tecnológico atual, tem sido fortemente evocado como premissa para a utilização de técnicas que auxiliem a exploração, análise e entendimento desses dados. Um conjunto dessas técnicas está compreendido na área de Visualização de Dados, que proporciona maneiras visuais de identificar padrões e tendências, além de extrair características obscuras dos dados. Tais abordagens podem ser aplicadas a qualquer problema que culmine na análise de dados. Um desses problemas é a utilização do som como ferramenta para descrever as características de uma paisagem, área denominada análise de Paisagens Acústicas. Nesta pesquisa de mestrado é apresentada uma abordagem visual para análise de Paisagens Acústicas. Essa abordagem consiste em duas etapas, sendo que na primeira são aplicadas técnicas visuais (Projeção Multidimensional t-SNE) e numéricas (Coeficiente de Silhueta) para avaliar quais conjuntos de características melhor descrevem uma Paisagem Acústica específica. A segunda etapa utiliza técnicas de visualização para analisar diferenças globais e características específicas de paisagens terrestres e aquáticas. Para isso foram utilizados o Mapa de Calor, as Coordenadas Paralelas e a xHiPP, uma extensão da projeção HiPP. A xHiPP busca melhorar a HiPP para alcançar melhor capacidade de análise e flexibilidade de aplicação. Com a aplicação das etapas apresentadas foi possível encontrar evidências de que o Mel-frequency Cepstrum Coefficients formam um conjunto de atributos eficaz para representação e segregação de Paisagens Acústicas. Também foi possível verificar que as técnicas de visualização empregadas na análise são capazes de destacar atributos semelhantes dos áudios, facilitando a análise, permitindo que o usuário dê enfoque às características relevantes do ambiente, no lugar de analisar áudios individuais para extrair informações.
The growth of data volume caused by current technological development has been strongly evoked as a premise for use of techniques that help exploration, analysis, and understanding of data. A subset of these techniques is yielded by the field of Data Visualization, which provides visual manners to identify patterns and trends, as well as the extraction of hidden data features. Such approaches can be applied to problems that aim at data analysis with a strong exploratory component. One such problem is the use of sound as a tool to describe environmental landscapes, named ecological Soundscapes. A visual approach to analysis Soundscapes is presented in this master research. The approach contains two steps, and the first step applies visual (tSNE Multidimensional Projection) and numeric (Silhouette Coefficient) techniques to evaluate attributes groups that better describe a specific Soundscape. The second step employs visual techniques to analysis global differences and specific features of the terrestrial and underwater environment. To achieve these goals, the research used Heatmap, Parallel Coordinates, and xHiPP, an extension of HiPP projection. The xHiPP enhanced HiPP to improve its analytical capabilities and flexibility. The presented steps were able to show evidence of the Mel-frequency Cepstrum Coefficients is an effective attribute collection to represent and segregate Soundscapes. As well, visual techniques employed in the analysis are capable to highlight similar audio features, making exploration easy, allowing users to focus relevant environmental attributes, instead of analyzing individual audios to extraction some information.

Soil-rock probing (Jb-probing) is the most common probing method in Sweden. Due to the penetration capacity of the Jb-probing it can be performed in both soil and rock. However, the capacity also results in inherent limitations and uncertainties, such as the difficulty identifying the soil layer sequences of soft soils. In order to attain a more detailed soil layer sequence it is necessary to perform complementary probing and sampling methods, an inefficient and consequently costly procedure. By instead implementing non-interfering complementary methods performed simultaneously as the Jb-probing the method may be rationalized. The so-called acoustic Jb-probing method may be a potential complement to the Jb-probing. In this thesis a continued study of the acoustic Jb-probing method is performed by means of a case study in Gubbängen with the focus on the potential additional information that the spectrogram (a visual representation of the frequency spectra) may contain compared to the Jb-parameters alone. This was done by obtaining vibration signals during Jb-probing using a triaxle geophone installed four meters from the boreholes. Vibration signals were collected from 13 boreholes. The vibration signals were then analyzed in time- and frequency domain which were compared to corresponding Jb-parameters and classified soil types. The results showed that the clay layers held the most promise for discovering additional information in the spectrogram, however this does not exclude potential in other soil types. Additionally, it was shown that the geophone ought to be fastened in the ground in order to attain satisfactory data. Overall, the acoustic Jb-probing method is a favorable way of collecting and analyzing data, which with continued development of the operational and computational process may be an economical alternative to the conventional method.

This project includes an overview of different processing methods phonograms. It focuses primarily on time - frequency analysis carried out using Fast Fourier Transform (STFT). The main essence is to evaluate the time - frequency analysis of recorded motor vehicle, driving at different options, together with an analysis of driver activity in these variants. Based on the results of the analysis deals with the creation of synthetic audio signals of motor vehicle records. Further addresses the creation of the program for acoustic simulation engine is running smoothly when driving a car. Processing recorded signals and their evaluation is transferred using Matlab 7.7.0 ( R2008 )

This thesis focuses on searching for CpG islands of DNA sequences based on analysis of DNA spectrograms. The first part is theoretical and deals with the significance CpG island, and a description of the algorithms that are used or have been proposed for their search. The theoretical basis were implemented two algorithms based on the analysis of DNA spectrogram. One is based on the assumption that the region CpG islands has a higher content of guanine and cytosine than the region outside the CpG island and the other on the assumption of a higher frequency of occurrence of CG dinucleotides in the CpG island. The algorithms are implemented through MATLAB programming interface. For evaluation usefulness and effectiveness of solutions, results achieved on the selected DNA sequences implemented algorithms are compared with the results achieved by search engines CpG islands, which are freely available on the internet.

In the past decade, the availability of digitally encoded, downloadable music has increased dramatically, pushed mainly by the release of the now famous MP3 compression format (Fraunhofer-Gesellschaft, 1994). Online sales of music in the US doubled in 2005, according to a recent news article (*), while the number of files exchanged on P2P platforms is much higher, but hard to estimate. The existing and coming informational flood in digital music prompts the need for sophisticated content-based information retrieval. Query-by-Humming is a prototypical technique aimed at locating pieces of music by melody; automatic annotation algorithms seek to enable finer search criteria, such as instruments, genre, or meter. Score transcription systems strive for an abstract, compressed form of a piece of music understandable by composers and musicians. Much research still has to be performed to achieve these goals. This thesis connects essential knowledge about music and human auditory perception with signal processing algorithms to solve the specific problem of pitch estimation. The designed algorithm obtains an estimate of the magnitude spectrum via STFT and models the harmonic structure of each pitch contained in the magnitude spectrum with Gaussian density mixtures, whose parameters are subsequently estimated via an Expectation-Maximization (EM) algorithm. Heuristics for EM initialization are formulated mathematically. The system is implemented in MATLAB, featuring a GUI that provides for visual (spectrogram) and numerical (console) verification of results. The algorithm is tested using an array of data ranging from single to triple superposed instrument recordings. Its advantages and limitations are discussed, and a brief outlook over potential future research is given. (*) "Online and Wireless Music Sales Tripled in 2005"; Associated Press; January 19, 2006

Snores are breath sounds that most people produce during sleep and they are reported to be a risk factor for various sleep disorders, such as obstructive sleep apnea syndrome (OSAS). Diagnosis of sleep disorders relies on the expertise of the clinician that inspects whole night polysomnography recordings. This inspection is time consuming and uncomfortable for the patient. There are surgical and therapeutic treatments. However, evaluation of the success of these methods also relies on subjective criteria and the expertise of the clinician. Thus, there is a strong need for a tool to analyze the snore sounds automatically, and to produce objective criteria and to assess the success of the applied treatment by comparing these criteria obtained before and after the treatment. In this thesis, we proposed a new algorithm to detect snoring episodes from the sleep sound recordings of the individuals, and created a user friendly interface to process snore recordings and to produce simple objective criteria to evaluate the results. The algorithm classifies sleep sound segments as snores and nonsnores according to their subband energy distributions. It was observed that inter- and intra-individual spectral energy distributions of snore sounds show significant similarities. This observation motivated the representation of the feature vectors in a lower dimensional space which was achieved using principal component analysis. Sleep sounds can be efficiently represented and classified as snore or nonsnore in a two dimensional space. The sound recordings were taken from patients that are suspected of OSAS pathology while they were connected to the polysomnography in Gü
lhane Military Medical Academy Sleep Studies Laboratory. The episodes taken from 30 subjects (18 simple snorers and 12 OSA patients) with different apnea/hypopnea indices were classified using the proposed algorithm. The system was tested by using the manual annotations of an ENT specialist as a reference. The system produced high detection rates both in simple snorers and OSA patients.

This diploma thesis deals with centrifugal pump running as a turbine. Basic working principles of a pump are included, both in pump and turbine regime. Experimental data obtained from laboratory test bed are compared with CFD simulation on slightly simplified geometry. Obtained results are then processed using spectrogram. Influence of time step and mesh size on results is also researched.

A big hurdle for many companies to start using machine learning is that trending techniques need a huge amount of structured data. One potential way to reduce the need for data is taking advantage of previous knowledge from a related task. This is so called transfer learning. A basic description of it would be when you take a model trained on existing data and reuse that for another problem. The purpose of this master thesis is to investigate if transfer learning can reduce the need for data when faced with a new machine learning task which is, in particular, to use transfer learning on ultrasound spectrograms of weld joints for predictive maintenance. The base for transfer learning is VGGish, a convolutional neural network model trained on audio samples collected from YouTube videos. The pre-trained weights are kept, and the prediction layer is replaced with a new prediction layer consisting of two neurons. The whole model is re-trained on the ultrasound spectrograms. The dataset is restricted to a minimum of ten and a maximum of 100 training samples. The results are evaluated and compared to a regular convolutional neural network trained on the same data. The results show that transfer learning improves the test accuracy compared to the regular convolutional neural network when the dataset is small. This thesis project concludes that transfer learning can reduce the need for data when faced with a new machine learning task. The results indicate that transfer learning could be useful in the industry.

For people with hearing impairments, attending to a single speaker in a multi-talker background can be very difficult and something which the current hearing aids can barely help with. Recent studies have shown that the audio stream a human focuses on can be found among the surrounding audio streams, using EEG and linear models. With this rises the possibility of using EEG to unconsciously control future hearing aids such that the attuned sounds get enhanced, while the rest are damped. For such hearing aids to be useful for every day usage it better be using something other than a motion sensitive, precisely placed EEG cap. This could possibly be archived by placing the electrodes together with the hearing aid in the ear. One of the leading hearing aid manufacturer Oticon and its research lab Erikholm Research Center have recorded an EEG data set of people listening to sentences and in which electrodes were placed in and closely around the ears. We have analyzed the data set by applying a range of signal processing approaches, mainly in the context of audio estimation from EEG. Two different types of linear sparse models based on L1-regularized least squares are formulated and evaluated, providing automatic dimensionality reduction in that they significantly reduce the number of channels needed. The first model is based on linear combinations of spectrograms and the second is based on linear temporal filtering. We have investigated the usefulness of the in-ear electrodes and found some positive indications. All models explored consider the in-ear electrodes to be the most important, or among the more important, of the 128 electrodes in the EEG cap.This could be a positive indication of the future possibility of using only electrodes in the ears for future hearing aids.

O presente relatório de estágio está divido em duas secções. A primeira aborda a realização do estágio na Escola de Música do Orfeão de Leiria – Conservatório de Artes, com o professor cooperante Nuno Mendes ao longo do ano letivo 2018/2019. A segunda parte deste relatório expõe a investigação, realizada ao longo da Prática de Ensino Supervisionada, sobre a utilização de palhetas sintéticas no ensino do saxofone. Inserido na família dos sopros de madeira, o saxofone retém uma série de problemáticas ligadas à sua ferramenta de produção de som, que mais comummente é a palheta feita de cana. Atualmente, as palhetas compostas por materiais sintéticos vêm resolver alguns dos aspetos negativos relacionados com o uso da madeira. Através deste estudo pretendeu-se analisar, usando métodos de gravação e análise de espetro sonoro, para observar os resultados obtidos e acompanhados ao longo de um ano letivo e o progresso de sujeitos de diferentes níveis de educação ao utilizar as palhetas sintéticas no seu quotidiano e em sala de aula; Report of Prática de Ensino Supervisionada realized in Escola de Música do Orfeão de Leiria – Adequation and Framework of the Use of Synthetic Reeds in Saxophone Teaching Abstract: The present internship report is divided in two sections. The first one addresses the internship done in Escola de Música do Orfeão de Leiria – Conservatório de Artes, with the cooperating teacher Nuno Mendes during the school year of 2018/2019. The second section of this report exposes the investigation, accomplished during the Prática de Ensino Supervisionada, about the use of synthetic reeds in saxofone teaching. Being in the woodwinds family, the saxophone has some of the problems associated to the use of the wooden reed as a mean of sound production. Currently, synthetic reeds are seen as a solution to some of the negative aspects that come with the use of the wooden reed. This research aims to comparatively evaluate the characteristics of synthetic reeds versus traditional cane reeds, through the use of audio recordings and spectrum analysis of the results obtained during the research period, as well as the progress made by the students while using synthetic reeds in a daily basis.

Низкочастотный (НЧ) шум (фликкер-шум, избыточный шум) присутствует во всех типах полупроводниковых приборов (ППП). Зависимость его спектральной плотности мощности (СПМ) от частоты имеет вид 1/fγ, где γ - показатель формы спектра. При измерении параметров НЧ-шума методом сравнения необходим генератор шума. Большинство серийных генераторов являются генераторами белого шума, в редких случаях – генераторами шума со спектром вида 1/f (т. е. с γ = 1).

This dissertation thesis deals with a new method for digital modulation recognition. The history and present state of the topic is summarized in the introduction. Present methods together with their characteristic properties are described. The recognition by means of artificial neural is presented in more detail. After setting the objective of the dissertation thesis, the digital modulations that were chosen for recognition are described theoretically. The modulations FSK, MSK, BPSK, QPSK, and QAM-16 are concerned. These modulations are mostly used in modern communication systems. The method designed is based on the analysis of module and phase spectrograms of the modulated signals. Their histograms are used for the examination of the spectrogram properties. They provide information on the count of carrier frequencies in the signal, which is used for the FSK and MSK recognition, and on the count of phase states on which the BPSK, QPSK, and QAM-16 are classified. The spectrograms in that the characteristic attributes of the modulations are visible are obtained with the segment length equal to the symbol length. It was found that it is possible to correctly recognize the modulation with the known symbol length at the signal-to-noise ratio at least 0 dB. That is why it is necessary to detect the symbol length prior to the spectrogram calculation. Four methods were designed for this purpose: autocorrelation function, cepstrum analysis, wavelet transform, and LPC coefficients. These methods were algorithmized and analyzed with signals disturbed by the white Gaussian noise, phase noise and with signals passed through a multipass fading channel. The method of detection by means of cepstrum analysis proved the most suitable and reliable. Finally the new method for digital modulation recognition was verified with signals passed through a channel with properties close to the real one.

Nonlinear dynamic analysis is a required step in seismic performance evaluation of many structures. Performing such an analysis requires input ground motions, which are often obtained through simulations, due to the lack of sufficient records representing a given scenario. As seismic ground motions are characterized by time-varying amplitude and frequency content, and the response of nonlinear structures is sensitive to the temporal variations in the seismic energy input, ground motion non-stationarities should be taken into account in simulations. This paper describes a nonparametric approach for modeling and prediction of non-stationary ground motions. Using Relevance Vector Machines, a regression model which takes as input a set of seismic predictors, and produces as output the expected evolutionary power spectral density, conditioned on the predictors. A demonstrative example is presented, where recorded and predicted ground motions are compared in time, frequency, and time-frequency domains. Analysis results indicate reasonable match between the recorded and predicted quantities.

Various methods of DNA sequences modifications for frequency analysis and basic characteristics of DNA are described in the theoretical part of this thesis. Tricolor spectrograms, created by short time Fourier transform help us to recognize some characteristic patterns in DNA sequences. Practical part of this work deals with developed programme which generates spectrograms and analyse them. Last part deals with the analysis of selected sequences of C. elegans genome. Some patterns are related to data of public databases such as NCBI. Various patterns are explained from the biological nature, which relates to chromosome structure and protein coding regions. Another well recognised patterns, tandem repetitions composed of satellites, microsatellites and minisatelites are described by spectrograms as well.

This thesis discusses about possibilities of construction colour DNA spectrograms and about patterns which can be detected there. Spectrograms as tools of spectral analysis give us a simultaneous view of the local frequency throughout the nucleotide sequence. They are suitable for gene identification or gene regions identification, determination of global character about whole chromosomes and also give us a chance for the discovery of yet unknown regions of potential significance. For purpose of this kind of DNA analysis is possible to use digital signal processing methods. We can apply them on only after conversion of DNA sequence to numerical representation. Selection of correct numerical representation affects how well will be reflected biological features in numerical record which we need for another use in digital signal analysis.

This work focuses on speech recognition by computer, pattern recognition stages and problems. Also there is a goal to create a speech recognition tool. At the beginning, there is a general overview of the audio signal and language concepts. The subsequent presentation of the essential tasks of speech recognition, introduction to matrix algebra, which is used to described algorithm. Information is provided on what basis and how features are extracted. For this work often is used LPC. This algorithm is one of the most popular extracting features of speech signal, so it is reviewed in this paper, as well as its modification WLPC. The following text of the speech recognition gives theory of extracted features use. Section „Acoustic modeling“ describes the recognition of speech units and one of the most commonly used acoustic modeling technologies – Hidden Markov Models and the next section „Speech modeling“ describes the language modeling, which purpose is to correct data referring to dictionaries and speech structure. The rest of the text is focused on speach recognition using specrtogram and implementation of speach recognition system. After that, there were executed experiments, that where used to define quality of speech recognition.
Šiame darbe gilinamasi i kalbos atpažinima kompiuteriu, atpažinimo etapus, problemas, o veliau meginama sukurti kalbos atpažinimo iranki. Pradžioje, bendrai apžvelgiama garso signalo, kalbos savokos. Veliau pateikiamos kalbos atpažinimo esminiai uždaviniai, supažindinama su matricu algebra, kuri naudojama aprašytuose algoritmuose. Pateikiama informacija kuo remiantis ir kaip išskiriami požymiai. Šiam darbui dažnai naudojamas LPC. Šis algoritmas yra vienas iš populiariausiu išskiriant kalbos signalo požymius, todel jis šiame darbe yra apžvelgtas, kaip ir jo modifikacija WLPC. Toliau tekste pateikiama kalbos atpažinimo teorija, apie išskirtu požymiu panaudojima. Skyriuje „Akustinis modeliavimas“, aprašomas kalbos vienetu atpažinimas ir vienas iš dažniausiai naudojamu akustinio modeliavimo technologiju - pasleptieji Markov’o modeliai, sekantis skyrius „Kalbos modeliavimas“, aprašo kalbos modeliavima, skirta jau turimiems duomenims sutvarkyti, remiantis žodynais ir analizuojamos kalbos struktura. Likusioje teksto dalyje koncentruojamasi ties kalbos atpažinimu panaudojant spektrograma ir kalbos atpažinimo sistemos igyvendinimu. Po to atlikti eksperimentai, kuriais buvo tiriama pateikto algoritmo atpažinimo kokybe.

Modern society has seen an increase in automation using AI in a variety of applications. To keep up with recent development, it is therefore logical to investigate the application of AI programs to military tasks. The great advantage with automation lies in the possible increase in efficiency and possible relocation of resources of personnel to other tasks. Therefore, this study aims to evaluate the use of Convolutional Neural Networks (CNN) in classification of communication and radar emitters based on collected Electronic Support Measures (ESM) data and to estimate to what extent human analysts could be replaced. The evaluation was performed by applying the concept of military Utility as a framework for evaluation with the addition of Technology Readiness Level (TRL) to survey how far the technology has developed. Data was collected using two methods: Firstly, through a literature review of research done on the application of CNNs in classifying information such as spectrograms and images. Secondly, by interviewing a subject matter expert from SAAB, who mainly helped estimate the TRL of the technology’s components. The study found that CNN appears suitable to apply on the proposed task and that the program could potentially replace human analysts to a great extent, at least when doing routine classifications. Full automation seems unlikely as analysts would be required with more challenging classifications, especially those outside the range of the training data used in teaching the CNN. Finally, challenges involved with deep learning programs inherent structure, demands and application to military tasks are discussed and subjects for future research are proposed.
Det moderna samhället har sett en ökad automatisering med AI i en mängd olika applikationer och för att hålla jämna steg med den senaste utvecklingen är det därför logiskt att undersöka tillämpningen av AI-program på militära uppgifter. Den stora fördelen med automatisering ligger i den möjliga ökningen av effektivitet och möjlig flytt av personalresurser till andra uppgifter. Därför syftar denna studie till att utvärdera användningen av convolutional neural networks (CNN) vid klassificering av kommunikations- och radarsändare baserat på insamlade data från elektronisk stödverksamhet (sv. ES motsvara eng. ESM) och att uppskatta i vilken utsträckning mänskliga analytiker kan ersättas. Utvärderingen genomfördes genom att använda konceptet militär nytta som ett ramverk för utvärdering med tillägg av technology readiness level (TRL) för att kartlägga hur långt tekniken har utvecklats. Data samlades in med två metoder: För det första genom en litteraturöversikt av forskning som gjorts om tillämpningen av CNN för att klassificera information såsom spektrogram och bilder. För det andra genom att intervjua en ämnesexpert från SAAB, som främst hjälpte till att uppskatta TRL för teknikens komponenter. Studien fann att CNN verkar lämplig att använda till den föreslagna uppgiften och att programmet potentiellt skulle kunna ersätta mänskliga analytiker i stor utsträckning, åtminstone for rutinklassificeringar. En fullständig automatisering verkar osannolik eftersom analytiker skulle krävas med mer utmanande klassificeringar, särskilt de som ligger utanför utbildningsdata som används för att lära upp programmet. Slutligen diskuteras utmaningar kopplade till djup-inlärningsprogrammens struktur, krav och tillämpning på militära uppgifter samt att ämnen för framtida forskning föreslås.

O estudo da penetração da radiação solar no meio aquático foi desenvolvido em quatro partes. Na primeira delas, voltada para a atenuação da radiação solar na coluna d’água, foram discutidos fatores de interferência e suas inter-relações. Neste sentido, foi feita uma comparação da penetração da radiação solar em ecossistemas continentais naturais e artificiais, analisando as diferenças e similaridades. O estudo comparativo da atenuação da radiação solar subaquática em toda extensão de reservatórios constituiu a segunda parte deste trabalho. A comparação foi feita entre reservatórios que apresentam proximidades de localização, mas com diferenças marcantes nos usos e ocupações nas regiões em que os mesmos estão inseridos. Na terceira parte foram apresentados dois recursos de grande utilização na obtenção das respostas imediatas do ambiente aquático, do ponto de vista óptico, diante dos impactos sofridos. O primeiro recurso foi a determinação do espectrograma da radiação solar, indicando a atenuação da radiação solar subaquática na faixa fotossinteticamente ativa (400 nm - 700 nm). O outro recurso foi a obtenção do espectro de absorção da água superficial, que permitiu o estabelecimento do “Índice de Absorção Espectral da Água”, I.A.E.A., que facilmente possibilita a detecção de fontes pontuais de poluição e contaminação. Como quarta e última parte foi proposto o desenvolvimento de um aparelho a partir da reconstrução do hidrofotômetro (Modelo Nº 268 WA 310 GM MFG & Instrument Corp. Bronx. N.Y. 10 451), onde foi introduzido e adaptado o detector fotocondutivo (PIN - 250DP) em substituição às fotocélulas anteriormente danificadas. Durante o trabalho de campo, este aparelho foi testado e foi feita a comparação com os obtidos do Quanta-meter LI-COR (LI - 185A) também utilizado neste trabalho.
The study of solar radiation penetration in aquatic environment was developed in four parts. The first one, directed to solar radiation attenuation in the water column, interference factors and their inter-relations were discussed. In this sense, a penetration comparison of the solar radiation in natural and artificial continental ecosystems was carried out, analyzing differences and similarities. The second part of this work consisted of a comparative study of underwater solar radiation in the entire reservoir. The comparison was done between reservoirs that are near to one another, but with marked differences in use and occupation within the regions where they are located. In the third part, two resources of great utilization in obtaining immediate response from the aquatic environment, from an optical point of view of the impacts experienced, were presented. The first resort was the spectrogram determination of solar radiation, which indicated the underwater solar radiation attenuation in the active photosynthetic range (400 nm - 700 nm). The other resort was the superficial water absorption spectrum, which allowed establishing the “Water Spectral Absorption Index”, I.A.E.A. that easily makes possible detection of accurate sources of pollution and contamination. The fourth and last part proposed to develop an apparatus from the construction of an underwater hydrophotometer (Model 268 WA 310 GM MFG & Instrument Corp. Bronx N.Y. 10451), where the photoconductor detector was adapted and introduced (PIN - 250DP) substituting the previously damaged photocells. During field work, this apparatus was tested and comparison was performed with data obtained from Quanta-meter LI-COR (LI - 185A), also used in this work.

With the breakthrough of machine learning techniques, the research concerning music emotion classification has been getting notable progress combining various audio features and state-of-the-art machine learning models. Still, it is known that the way to preprocess music samples and to choose which machine classification algorithm to use depends on data sets and the objective of each project work. The collaborating company of this thesis, Ichigoichie AB, is currently developing a system to categorize music data into positive/negative classes. To enhance the accuracy of the existing system, this project aims to figure out the best model through experiments with six audio features (Mel spectrogram, MFCC, HPSS, Onset, CENS, Tonnetz) and several machine learning models including deep neural network models for the classification task. For each model, hyperparameter tuning is performed and the model evaluation is carried out according to pareto optimality with regard to accuracy and execution time. The results show that the most promising model accomplished 95% correct classification with an execution time of less than 15 seconds.
I och med genombrottet av maskininlärningstekniker har forskning kring känsloklassificering i musik sett betydande framsteg genom att kombinera olikamusikanalysverktyg med nya maskinlärningsmodeller. Trots detta är hur man förbehandlar ljuddatat och valet av vilken maskinklassificeringsalgoritm som ska tillämpas beroende på vilken typ av data man arbetar med samt målet med projektet. Denna uppsats samarbetspartner, Ichigoichie AB, utvecklar för närvarande ett system för att kategorisera musikdata enligt positiva och negativa känslor. För att höja systemets noggrannhet är målet med denna uppsats att experimentellt hitta bästa modellen baserat på sex musik-egenskaper (Mel-spektrogram, MFCC, HPSS, Onset, CENS samt Tonnetz) och ett antal olika maskininlärningsmodeller, inklusive Deep Learning-modeller. Varje modell hyperparameteroptimeras och utvärderas enligt paretooptimalitet med hänsyn till noggrannhet och beräkningstid. Resultaten visar att den mest lovande modellen uppnådde 95% korrekt klassificering med en beräkningstid på mindre än 15 sekunder.

Cette thèse explore l'application de la Morphologie Mathématique à l'analyse et la génération de musique, en se concentrant sur deux représentations en temps-fréquence : les spectrogrammes et les piano rolls. La morphologie mathématique est un outil de traitement des images non linéaire qui sert à considérer des notions topologiques de l'image. Nous présentons trois applications. La première est d'analyser des spectrogrammes avec des outils morphologiques pour obtenir des paramètres avec lesquels synthétiser un son d'instrument de musique. La deuxième est de générer des piano rolls avec deux paramètres musicaux, la texture et l'harmonie, en les disposant grâce à la dilatation morphologique. La troisième est d'appliquer des opérateurs morphologiques pour analyser des piano roll à l'aide de la théorie des graphes. La thèse propose donc de nouvelles approches pour des problèmes d'analyse sonore et musicologie computationnelle
This thesis explores the application of Mathematical Morphology to the analysis and generation of music, focusing on two time-frequency representations: spectrograms and piano rolls. Mathematical Morphology is a nonlinear image processing tool that serves to consider topological notions of the image. We present three applications. The first is to analyze spectrograms with morphological tools to obtain parameters with which to synthesize a musical instrument sound. The second is to generate piano rolls with two musical parameters, texture and harmony, by arranging them through morphological dilation. The third is to apply morphological operators to analyze piano rolls using graph theory. The thesis thus proposes new approaches for problems in sound analysis and computational musicology

This thesis investigates issues of intelligibility through the lens and focus of prosody when the Bostonian and Los Angeles-based accents are heard in casually occurring conversation by native and non-native speakers. Over the spring and summer of 2017, six native speakers and 11 non-native speakers of English were interviewed from having listened to two 2.5 minute audio sample clips of speakers who have these accents. Respondents were asked questions such as what was difficult or easy or whether they could summarize the recordings for me. Findings indicate that while the native speakers often had difficulty with vocabulary due to context, non-native English speakers frequently found the same recordings to sound continuous, blended or merged together when the Boston and Los Angeles audios were played to them. Native English speakers, by contrast, did not seem to face the same prosodic challenges of intelligibility as their non-native English-speaking counterparts when these two accents were heard in informal conversation. It has been found that L2 learners have a strong desire to learn English from their teachers through more naturally or informally occurring conversation. The argument is made that the teaching, practice and engagement of informal conversation is woefully inadequate for non-native speakers of English. Within this thesis the core subsets of the perceptions of prosody are analyzed between native and non-native speakers of English. The purpose of doing so is to pedagogically improve learning in EFL and ESL contexts.

SALT HRS (Southern African Large Telescope High Resolution Échelle Spectrograph) is a high-resolution, high-efficiency spectrograph for the 11m SALT telescope in Sutherland, South Africa. The initial optical design work was performed at the University of Canterbury, New Zealand. Revisions to the concept, the mechanical design, manufacture, assembly and testing have been handled by the Centre for Advanced Instrumentation, at Durham University in the United Kingdom. SALT HRS is a fibre-fed echelle grating spectrograph with four operational modes: low-, medium- and high-resolution and high-stability modes, having spectral resolutions of R ≈16000, 37000, 67000 and 67000 respectively over a wavelength range of 370-890nm. The instrument is of a dual channel, ‘white pupil’ design, in which the primary mirror acts to collimate light onto a single R4 echelle grating, and also to focus dispersed light to an intermediate focus. A dichroic beam-splitter separates the dispersed light into two separate spectral channels. Spherical pupil mirrors transfer the separated beams via a fold mirror to two wavelength-specific volume-phase holographic gratings (VPHGs) used as cross-dispersers. Cross-dispersed spectra are then imaged by two fully dioptric camera systems onto optimized CCD detectors. This thesis presents the results of the laboratory testing and specification of several critical sub-systems of SALT HRS, as well as the development of key software tools for the design verification and operation at the telescope. In Chapter 1 we first review the technical development of high-resolution spectroscopy and its specific implementation in SALT HRS. In Chapter 2 we develop a comprehensive throughput model of the entire system based on a combination of as-built performance and specific throughput measurements in the laboratory. This is used to make some specific predictions for the on-sky performance of SALT HRS and the magnitude limits for science targets. We also present a graphical exposure time calculator based on these measurements which can be used by an astronomer to plan their observations with SALT HRS. Chapter 3 contains a detailed treatise on the optical fibre system of SALT HRS. Considerations for the use of optical fibres in astronomy are provided, as are details of an optional double scrambler, and the various instrument fibre modes. Extensive measurements of focal ratio degradation (FRD) are also presented, with testing of input beam speed; wavelength; fibre bending; variable pupil mirror illumination; and vacuum tank pressure dependency. The systems for fibre management are reviewed, as is the fibre bundle assembly process. Testing of two further sub-systems is described in Chapter 4. Firstly the long-term stability of the mirror mounting mechanisms is determined. The advantages of cross-dispersion of echelle spectra using volume-phase holographic gratings are then discussed, and the results of diffraction efficiency measurements are given for both red and blue channel gratings. Modern CCD technologies are examined in Chapter 5, and the blue detector is experimentally characterized using photon transfer and quantum efficiency curves. It is also used for an investigation into cosmic ray events in CCDs. Results from shielding the detector using lead are described, as is an attempt to distinguish the source of the events based on their morphology. Finally, Chapter 6 deals with the handling of data produced by SALT HRS. Methods of wavelength calibration of the spectra are discussed, including the use of Thorium-Argon lamps and an iodine absorption cell. The implementation of a Python based quick-look data reduction pipeline is reviewed, with a description of the processes performed. A summary of the thesis is given in Chapter 7.

Brushless Direct Current (BLDC) motors are one of the motor types rapidly gaining popularity. BLDC motors are being increasingly used in critical high performance industries such as appliances, automotive, aerospace, consumer, medical, industrial automation equipment and instrumentation. Fault detection and condition monitoring of BLDC machines is therefore assuming a new importance. The objective of this research is to advance the field of rotor and load fault diagnosis in BLDC machines operating in a variety of operating conditions ranging from constant speed to continuous transient operation. This objective is addressed as three parts in this research. The first part experimentally characterizes the effects of rotor faults in the stator current and voltage of the BLDC motor. This helps in better understanding the behavior of rotor defects in BLDC motors. The second part develops methods to detect faults in loads coupled to BLDC motors by monitoring the stator current. As most BLDC applications involve non-stationary operating conditions, the diagnosis of rotor faults in non-stationary conditions forms the third and most important part of this research. Several signal processing techniques are reviewed to analyze non-stationary signals. Three new algorithms are proposed that can track and detect rotor faults in non-stationary or transient current signals.

O presente estudo tem como objetivo determinar se a gravidade da DPOC poderá ser estimada através da área do gráfico das intensidades sonoras dos sons respiratórios de pacientes com DPOC. O estudo realizado com 51 pacientes com DPOC leve, moderado, grave, muito grave e 7 indivíduos saudáveis não fumantes. Os sons respiratórios de cada participante, foram coletados através de estetoscópio adaptado com um mini microfone. O método compara as áreas das intensidades sonoras em função da frequência de pacientes de DPOC e indivíduos saudáveis. Neste contexto, para atender ao objetivo, um método foi proposto e testado baseado na combinação de técnicas de filtragem e TFTC, seguida de análise estatística, cálculo da média, desvio padrão e interpolação. Os resultados sugerem que a área do gráfico da variância da intensidade sonora em função da frequência diminui quando aumenta a gravidade da DPOC, exceto para os casos em que a bronquite crônica é predominante.
The present study aims to determine if the severity of COPD can be estimated through the chart area of the sound intensities of respiratory sounds in patients with COPD. The study included 51 patients with mild, moderate, severe, very severe COPD and 7 healthy non-smokers. The breathing sounds of each participant were collected through a stethoscope adapted with a mini microphone. The method compares the areas of intensity sonic densities as a function of the frequency of COPD patients and healthy individuals. In this context, to meet the objective, a method was proposed and tested based on the combination of filtering techniques and TFTC, followed by statistical analysis, calculation of the mean, standard deviation and interpolation. The results suggest that the area of the graph of frequency-frequency sound intensity variance decreases as the severity of COPD increases, except for cases where chronic bronchitis is predominant.

Nos travaux de recherches présentés dans ce manuscrit ont pour objectif, l'optimisation des performances des algorithmes de suivi des formants. Pour ce faire, nous avons commencé par l'analyse des différentes techniques existantes utilisées dans le suivi automatique des formants. Cette analyse nous a permis de constater que l'estimation automatique des formants reste délicate malgré l'emploi de diverses techniques complexes. Vue la non disponibilité des bases de données de référence en langue arabe, nous avons élaboré un corpus phonétiquement équilibré en langue arabe tout en élaborant un étiquetage manuel phonétique et formantique. Ensuite, nous avons présenté nos deux nouvelles approches de suivi de formants dont la première est basée sur l'estimation des crêtes de Fourier (maxima de spectrogramme) ou des crêtes d'ondelettes (maxima de scalogramme) en utilisant comme contrainte de suivi le calcul de centre de gravité de la combinaison des fréquences candidates pour chaque formant, tandis que la deuxième approche de suivi est basée sur la programmation dynamique combinée avec le filtrage de Kalman. Finalement, nous avons fait une étude exploratrice en utilisant notre corpus étiqueté manuellement comme référence pour évaluer quantitativement nos deux nouvelles approches par rapport à d'autres méthodes automatiques de suivi de formants. Nous avons testé la première approche par détection des crêtes ondelette, utilisant le calcul de centre de gravité, sur des signaux synthétiques ensuite sur des signaux réels de notre corpus étiqueté en testant trois types d'ondelettes complexes (CMOR, SHAN et FBSP). Suite à ces différents tests, il apparaît que le suivi de formants et la résolution des scalogrammes donnés par les ondelettes CMOR et FBSP sont meilleurs qu'avec l'ondelette SHAN. Afin d'évaluer quantitativement nos deux approches, nous avons calculé la différence moyenne absolue et l'écart type normalisée. Nous avons fait plusieurs tests avec différents locuteurs (masculins et féminins) sur les différentes voyelles longues et courtes et la parole continue en prenant les signaux étiquetés issus de la base élaborée comme référence. Les résultats de suivi ont été ensuite comparés à ceux de la méthode par crêtes de Fourier en utilisant le calcul de centre de gravité, de l'analyse LPC combinée à des bancs de filtres de Mustafa Kamran et de l'analyse LPC dans le logiciel Praat. D'après les résultats obtenus sur les voyelles /a/ et /A/, nous avons constaté que le suivi fait par la méthode ondelette avec CMOR est globalement meilleur que celui des autres méthodes Praat et Fourier. Cette méthode donne donc un suivi de formants (F1, F2 et F3) pertinent et plus proche de suivi référence. Les résultats des méthodes Fourier et ondelette sont très proches dans certains cas puisque toutes les deux présentent moins d'erreurs que la méthode Praat pour les cinq locuteurs masculins ce qui n'est pas le cas pour les autres voyelles où il y a des erreurs qui se présentent parfois sur F2 et parfois sur F3. D'après les résultats obtenus sur la parole continue, nous avons constaté que dans le cas des locuteurs masculins, les résultats des deux nouvelles approches sont notamment meilleurs que ceux de la méthode LPC de Mustafa Kamran et ceux de Praat même si elles présentent souvent quelques erreurs sur F3. Elles sont aussi très proches de la méthode par détection de crêtes de Fourier utilisant le calcul de centre de gravité. Les résultats obtenus dans le cas des locutrices féminins confirment la tendance observée sur les locuteurs
Our research work presented in this thesis aims the optimization of the performance of formant tracking algorithms. We began by analyzing different existing techniques used in the automatic formant tracking. This analysis showed that the automatic formant estimation remains difficult despite the use of complex techniques. For the non-availability of database as reference in Arabic, we have developed a phonetically balanced corpus in Arabic while developing a manual phonetic and formant tracking labeling. Then we presented our two new automatic formant tracking approaches which are based on the estimation of Fourier ridges (local maxima of spectrogram) or wavelet ridges (local maxima of scalogram) using as a tracking constraint the calculation of center of gravity of a set of candidate frequencies for each formant, while the second tracking approach is based on dynamic programming combined with Kalman filtering. Finally, we made an exploratory study using manually labeled corpus as a reference to quantify our two new approaches compared to other automatic formant tracking methods. We tested the first approach based on wavelet ridges detection, using the calculation of the center of gravity on synthetic signals and then on real signals issued from our database by testing three types of complex wavelets (CMOR, SHAN and FBSP). Following these tests, it appears that formant tracking and scalogram resolution given by CMOR and FBSP wavelets are better than the SHAN wavelet. To quantitatively evaluate our two approaches, we calculated the absolute difference average and standard deviation. We made several tests with different speakers (male and female) on various long and short vowels and continuous speech signals issued from our database using it as a reference. The formant tracking results are compared to those of Fourier ridges method calculating the center of gravity, LPC analysis combined with filter banks method of Kamran.M and LPC analysis integrated in Praat software. According to the results of the vowels / a / and / A /, we found that formant tracking by the method with wavelet CMOR is generally better than other methods. Therefore, this method provides a correct formant tracking (F1, F2 and F3) and closer to the reference. The results of Fourier and wavelet methods are very similar in some cases since both have fewer errors than the method Praat. These results are proven for the five male speakers which is not the case for the other vowels where there are some errors which are present sometimes in F2 and sometimes in F3. According to the results obtained on continuous speech, we found that in the case of male speakers, the result of both approaches are particularly better than those of Kamran.M method and those of Praat even if they are often few errors in F3. They are also very close to the Fourier ridges method using the calculation of center of gravity. The results obtained in the case of female speakers confirm the trend observed over the male speakers

Voice analysis is an important tool in the diagnosis of laryngeal disorders. Among distinct signal processing techniques employed for voice analysis, the spectrogram is commonly used, as it allows for a visualization of the variation of the energy of the signal as a function of the both time and frequency. In this context, this study investigates the use of the global energy of the voice signal, estimated through the spectrogram, as a tool for discrimination between signals obtained from healthy and pathological subjects. This research has also exploited the potential use of the global energy of the voice signal to discriminate distinct laryngeal disorders. In total, 94 subjets were involved in this study, from which 46 were dysphonic and 48 normal. The diagnosis of laryngeal disorders was confirmed by means of a videolaryngoscopic examination. Participants were also subjected to a clinical examination of vocal acoustic through the recording of the sustained vowel. The global energy allowed for the discrimination between normal and disphonic voice. Furthermore, this technique could discriminate the voice signal of patients suffering from left vocal fold paralysis from those suffering from other investigated disorders. The results suggest the global energy of the signal as an auxiliary and alternative tool for the diagnosis between normal and dysphonic voice.
A análise do sinal de voz é uma ferramenta importante no diagnóstico dos distúrbios laríngeos. Dentre as diversas técnicas para o processamento da voz destaca-se o espectrograma por permitir uma visualização da variação da energia do sinal em função do tempo e freqüência. Neste contexto, esta pesquisa investiga a energia global do sinal de voz, estimada a partir do espectrograma, como ferramenta capaz de discriminar esses sinais, obtidos de pacientes com diferentes doenças, daqueles coletados de sujeitos saudáveis. O estudo ainda verifica a possibilidade do uso da energia global na discriminação de distúrbios laríngeos. No total 94 indivíduos participaram desse estudo, sendo 46 disfônicos e 48 eufônicos. Inicialmente os sujeitos foram submetidos a um exame de videolaringoscopia para a determinaçãoção do diagnóstico. Posteriormente eles foram submetidos a um exame clínico de análise acústica vocal por meio da gravação da vogal sustentada. Utilizando a energia global foi possível discriminar as vozes normais das vozes disfônicas. Por meio da energia global foi possível ainda separar as vozes de pacientes portadores de paralisia da prega vocal esquerda de todas as outras doenças laríngeas investigadas. Podemos sugerir a energia global da voz como uma ferramenta auxiliar no diagnóstico diferencial entre vozes normais e disfônicas.
Mestre em Ciências

The awareness of the human responsibility in global warming has led to various private and public initiatives to reduce the emission of greenhouse gases, up to international level. In this context the development of renewable technologies in two sectors having an important ecological footprint, i.e. production of electricity and transportation, is targeted.In the firstly mentioned sector, the progression of wind energy is at present the most rapid among all renewable energies. But wind turbines still suffer from a global lack of reliability and accessibility compared to classical power plants, leading to potentially important production losses and repair costs. The first part of the present work focuses on the improvement of the electrical chain reliability by the combination of an estimator and a fault-detection algorithm to achieve sensor-fault tolerance, taking benefit from the already available measurement redundancies on doubly-fed-induction-machine (DFIG) drives.Estimators and sensor-fault detection and isolation (FDI) in DFIGs have been the object of many research papers. However, most of them only consider one unique type of measurement and only a few works consider magnetic saturation. A new combination of a closed-loop observer with a cumulative-sum-based FDI technique, considering magnetic saturation and using limited computational resources is proposed here to estimate electromagnetic torque, rotor currents and position for sensor-fault detection and tolerance. This algorithm is then validated in steady state and in case of moderate transients, unbalanced conditions and misestimation of DFIG parameters. The estimator can also start on the fly during the start-up process of the generator.In the transportation sector, new hybrid and full-electric vehicles start to be visible on the roads, but still need important technological improvements in terms of autonomy, performances, but also produced noise and vibrations. The objectives of the second part of this doctoral thesis are related to this last challenge and consist of the experimental investigation of noise, vibration and harshness (NVH) aspects of an 8/6 switched-reluctance machine (SRM) designed for an electrical vehicle (EV).The NVH issues of SRMs, limiting their usage in automotive and other domains, have been the subject of various papers. However, most of them focus on modal analysis or detailed phenomena, while a global evaluation of NVH aspects of SRMs in normal working conditions is rarely made, as well as the use of reproducible sound metrics. A global and relatively fast experimental method to assess the evolution of noise and vibration is proposed. Tests are performed in transient regime, using continuously varying working conditions when possible, for the excitation of a large band of frequencies. The resulting current, radial vibration and acoustic noise are presented as spectrograms for an easy distinction of affected and unaffected frequencies and compared with the associated loudness and sharpness.Furthermore, the implementation of a new faster-sampled current-hysteresis controller has allowed to improve the quality of the control and of the acoustic noise by reducing the current-ripple amplitude and the excitation of resonances. The various tests show that the switching frequency has to be high enough to avoid exciting the ovalization mode of the SRM, but not too high to avoid producing a too sharp noise. The ripple amplitude also has to be considered to limit the loudness. Therefore, soft chopping, or a reduced DC-bus voltage at low speeds, has to be preferred with a relative small hysteresis bandwidth. Finally, the case of an open-phase fault has been investigated showing amplified even current orders in the vibration and acoustic-noise plots.
La prise de conscience de la responsabilité humaine dans le réchauffement climatique est à la source de nombreuses initiatives publiques et privées parfois internationales pour réduire les émissions de gaz à effet de serre. Dans ce contexte, le développement de technologies durables dans deux secteurs à forte empreinte écologique est visé: la production d'énergie électrique et les transports. Dans le premier secteur, la progression de l'éolien est à présent la plus rapide parmi toutes les énergies renouvelables. Cependant, les éoliennes souffrent d'un manque global de fiabilité et d'accessibilité par rapport aux centrales électriques classiques, ce qui conduit potentiellement à des pertes de production et des coûts de réparation importants. La première partie de ce travail se focalise sur l'amélioration de la chaîne électrique en la rendant tolérante aux défauts de capteurs au moyen de la combinaison d'un estimateur et d'un algorithme de détection de défauts, tirant avantage de la redondance de mesures déjà présente sur les entraînements à machines asynchrones à double alimentation (MADA). Les estimateurs et la détection et l'isolation de défauts de capteurs sur les MADA a fait l'objet de nombreuses publications scientifiques. Cependant, la plupart d'entre elles considèrent un seul type de mesure et peu de travaux prennent en compte la saturation magnétique. Une nouvelle combinaison d'un observateur et d'un algorithme de détection de défauts de type ‘CUSUM', considérant la saturation magnétique et nécessitant une puissance de calcul limitée, est proposée dans cette thèse pour l'estimation du couple électromagnétique, des courants et de la position rotoriques en vue d'obtenir la tolérance aux défauts de capteurs. Cet algorithme est validé en régime permanent et cas de transitoires modérés, de tensions du réseau déséquilibrées et d'erreurs d'estimation des paramètres de laMADA. L'estimateur est aussi capable de démarrer seul lors du démarrage de la génératrice. Dans le secteur des transports, des véhicules hybrides et électriques commencent à être visibles sur les routes, malgré que des progrès technologiques importants en termes d'autonomie, de performances, mais aussi de bruits et vibrations soient encore nécessaires pour une utilisation plus intensive. L'objectif de la deuxième partie de cette thèse se rapporte à ce dernier défi et consiste à analyser les aspects acoustiques et vibratoires d'une machine à réluctance variable 8/6 conçue pour propulser un véhicule électrique. Ces problèmes acoustiques et vibratoires, qui limitent notamment l'usage de telles machines dans des applications de propulsion, ont été l'objet de divers articles scientifiques. Cependant, la plupart d'entre eux sont focalisés sur des analyses modales ou de phénomènes particuliers, alors qu'une évaluation globale des problèmes de bruit et de vibration des machines à réluctance variable en conditions normales de fonctionnement est rarement proposée, de même que l'utilisation de critères de qualité sonore. Une méthode expérimentale globale et relativement rapide pour évaluer l'évolution du bruit et des vibrations est proposée dans ce travail. Les essais sont réalisés en régime transitoire pour exciter une large bande de fréquences et en faisant varier continuellement, quand cela est possible, les conditions de fonctionnement. Les courants, vibrations radiales et bruits acoustiques résultants sont présentés sous formes de cartographies couleur pour une distinction aisée des fréquences affectées et non-affectées et comparés aux niveaux calculés de bruyance et d'acuité correspondants. Par ailleurs, la mise en place d'un nouveau régulateur à hystérèse en courant à plus grande fréquence d'échantillonnage a permis d'améliorer la qualité de la commande et du bruit acoustique associé en réduisant l'amplitude des oscillations de courant et l'excitation des fréquences de résonance. Les essais montrent que la fréquence de commutation doit être suffisamment élevée pour éviter l'excitation du mode d'ovalisation de la machine, mais pas trop pour éviter une trop grande acuité du son produit. L'amplitude des oscillations doit aussi être considérée pour limiter la bruyance. En conséquence, une commande en ‘soft chopping', ou une tension réduite du bus continu à basse vitesse, doit être combinée à une bande d'hystérèse relativement faible. Enfin, le cas d'un défaut de phase ouverte a été étudié et a montré une amplification des ordres pairs du courant dans les spectres vibratoires et acoustiques.
De bewustwording van de menselijke verantwoordelijkheid in de opwarming van de aarde heeft tot verschillende private en publieke initiatieven geleid om de uitstoot van broeikasgassen te verminderen. In deze context is de ontwikkeling van hernieuwbare technologieën hoofdzakelijk gericht op twee sectoren met een belangrijke ecologische impact: elektriciteitsproductie en transport.In de eerste sector ontwikkelt windenergie zich op dit moment sneller dan alle andere hernieuwbare energieën. Maar windturbines lijden nog steeds aan een gebrek aan betrouwbaarheid en toegankelijkheid, en dus aan potentieel hogere productieverliezen en herstelkosten, als ze met klassieke krachtcentrales worden vergeleken. In het eerste deel van deze doctoraatsthesis wordt op de verbetering van de elektrische keten geconcentreerd door de combinatie van een schatter en een foutdetectie- en -isolatiealgoritme (FDI-algoritme) om sensorfouttolerantie te verkrijgen dankzij de reeds aanwezige meetovertolligheid op dubbelgevoede inductiemachine (DFIG) aandrijvingen.Schatters en sensor-FDI-algoritmen zijn het onderwerp van vele wetenschappelijke artikelen geweest. Meestal wordt maar één sensortype beschouwd en met de magnetische verzadiging wordt niet vaak rekening gehouden. Een nieuwe combinatie van een schatter met gesloten terugkoppeling en een FDI-techniek gebaseerd op het ‘cumulative-sum' principe is voorgesteld. Zo kan het elektromagnetische koppel, de rotorstromen en positie worden geschat voor sensor FDI en fouttolerantie met beperkte rekenkosten en zonder de magnetische verzadering te verwaarlozen. Het algoritme wordt in stabiele toestand gevalideerd, maar ook in het geval van gematigde transiënte situaties, onevenwichtige netwerkomstandigheden en een verkeerde schatting van DFIG parameters. Het kan ook vanzelf starten tijdens de startprocedure van de generator.In de vervoersector beginnen hybride en elektrische voertuigen op de wegen te rijden. Maar vooreen intensiever gebruik van zo'n wagens zijn er nog technologische verbeteringen nodig met betrekking tot autonomie, prestaties en ook geluid en trillingen (NVH). Het tweede deel van de thesis betreft die laatste uitdaging en bestaat uit het experimentele onderzoek van geluid en trillingen op een 8/6 variabelereluctantiemachine (SRM) ontwikkeld voor elektrische voertuigen.De NVH-problemen van SRM's beperken hun gebruik in automobiele en andere toepassingen enonderzoek wordt erover voortgezet. Vele wetenschappelijke artikelen focussen toch op modale analyse of gedetailleerde fenomenen terwijl een globale evaluatie van NVH aspecten in SRM's in gewone operatiecondities nauwelijks wordt gemaakt. Hetzelfde geldt voor het gebruik van reproduceerbare geluidsmetrieken. Een globale en vrij vlugge experimentele methode is hier voorgesteld om het NVH gedrag te schatten. Testen worden in transiënte situaties uitgevoerd om een brede frequentieband te exciteren, indien mogelijk met voortdurend variërende condities. De gemeten fasestroom, trilling en geluid worden als kleurmappen geplot om het verschil tussen beïnvloede en niet geaffecteerde frequenties te vergemakkelijken en met de berekende akoestische luidheid en scherpte vergeleken.Bovendien heeft de implementatie van een sneller bemonsterd stroomhysteresisregelaar geleid tot een verbetering van de regulatie- en akoestische kwaliteit door de amplitude van de stroomrimpeling en de excitatie van resonantiefrequenties te verminderen. De testresultaten tonen dat de schakelfrequentie voldoende hoog moet zijn om de excitatie van de ovale vervormingsmode te vermijden, maar niet te hoog om de scherpte van het geluid te beperken. De amplitude van de rimpel beïnvloedt ook de luidheid en daarvoor moet in aanmerking worden genomen. Bijgevolg zou ‘soft chopping'mode, of een lagere spanning op de DC-bus bij lage toerentallen, met een relatief klein hysteresisband beter worden gebruikt. Uiteindelijk wordt het geval van een openfasefout bestudeerd en onthult versterkte gelijke frequentievolgorden in de trilling- en geluidplots.
Doctorat en Sciences de l'ingénieur et technologie
info:eu-repo/semantics/nonPublished

L'obiettivo della presente tesi è di introdurre una nuova concezione di fotorivelatore astrofisico, in grado di analizzare immediatamente la composizione spettrale di immagini bidimensionali. Prima di tutto, la sua innovativa tecnologia, che si basa sull'effetto superconduttivo dell'induttanza cinetica, è confrontata con lo standard dei dispositivi semiconduttivi ad accoppiamento di carica. Essi contano sull'effetto fotoelettrico e possono solo percepire l'intensità della luce su un periodo d'esposizione, di per sé, senza elementi dispersivi o filtri monocromatici dinnanzi. In questo contesto, riporto il risultato della mia collaborazione con l'Università di Oxford, dov'è in via di sviluppo uno spettrografo a campo integrale chiamato KIDSpec, operativo nella banda elettromagnetica che va dall'ultravioletto al vicino infrarosso. Sarà il primo strumento di tal sorta ad essere costruito al di fuori degli Stati Uniti d'America. Il mio compito consisteva nell'assemblare uno spettrometro ottico destinato ad uno studio di fattibilità di questo apparato. Quindi, le carenze del sistema ottico sono discusse in breve, in vista della sua implementazione. A seguire, le indagini ottiche attraverso cui esaminai la prestazione dello spettrometro, quanto alle sue risposte in fatto di diffrazione, a varî angoli d'uscita e lunghezze d'onda in ingresso. In aggiunta, si fornisce un'eziologia dei motivi circolari causati dalla fibra ottica ivi impiegata. Infine, sono esaminate alcune possibili applicazioni astrofisiche del nuovo tipo di rivelatore: particolare attenzione è data all'emergente frontiera della caratterizzazione esoplanetaria diretta.

This thesis describes an aspect of the operational and functional implementation of the CORES instrument, an electronion spectrograph to be flown aboard the Russian Obstanovka Experiment Complex mission to the ISS in 2008. For the phenomenon under investigation, wave-particle interactions in space plasma, CORES is ideally suited as it can accept a very wide band of energies simultaneously and at the same time have 3600 field-of-view. This instrument includes a novel delay-baSed particle readout system, . developed specifically for this application. A reconfigurable device (FPGA) lies at the heart of the instrument, and this has been taken fully advantage of to achieve the necessary functionality. As such this thesis mostly focuses on the configuration of this device, which allows for the multi-channel particle capturing, as well as defining the processing abilities of the instrument. These include a number of novel functions implemented in hardware for spectral analysis of the recorded particle events, such as a parallel version of se~ond order correlation, a parallel version of the I-bit autocorrelation function, etc. Furthermore, an innovative tool for on-board telemetry filtering is also presented, namely the on-board calculation of the information content of the collected measurements. This thesis also offers a technical view of CORES, presenting its capabilities based on its hardware structure.

The Cambridge Infrared Panoramic Survey Spectrograph, CIRPASS, is an infrared spectrograph that operates in the 0.9 to 1.8 mm wavelength region. CIRPASS was developed entirely at the Institute of Astronomy, Cambridge, and the design, construction and performance of CIRPASS are presented in this thesis. CIRPASS provides integral field spectroscopy (IFS). Spectra are simultaneously obtained from 499 contiguous spatial elements in a two-dimensional area of sky using an integral field unit (IFU). The construction of the IFU, which used a lenslet array and fibre optic bundle, is presented, as well as a review of IFS techniques. The sensitivity of CIRPASS has been maximised by ensuring the background signal detected by the instrument is reduced to an absolute minimum. CIRPASS has the capability to remove, both in hardware and software, the dominant infrared sky background from OH emission in the upper atmosphere. The advantages that this offers, and the benefits of both suppression techniques are presented. The thermal background of CIRPASS was reduced by cooling the entire spectrograph to -40°C. The instrument was cooled using a conventional industrial refrigeration system and resulted in significant improvements to the instrument's performance. The characteristics of the HAWAII infrared detector array used were measured, to determine their impact on the instrument's sensitivity, and found to be within specification. Software was written to reduce CIRPASS data that is applicable to any IFS instrument producing data in the Euro 3D format. A data reduction pipeline, that can be used for rapid data reduction at the telescope and for more detailed analysis afterwards, was implemented. The pipeline implementation and the techniques adopted for reducing closely packed spectral data on an infrared detector are discussed. The performance of CIRPASS was quantified by observing the Cambridge night sky with CIRPASS attached to a small telescope. The results provide an accurate measure of the instrument's performance on a larger telescope and are detailed. The expected limiting magnitudes for 3 hour observations of a point source, on an 8 m telescope, resulting in a signal-to-noise ratio of 5 are J=21.3 and H=19.3. The equivalent limiting line fluxes are J=4.4x10^-18 ergs/s/cm² and H=9.0x10^-8 ergs/s/cm².

Prior speech and linguistics research has focused on the use of phonemes recognition in speech, and their use in formulation of recognizable words, to determine language identification. Some languages have additional phoneme sounds, which can help identify a language; however, most of the phonemes are common to a wide variety of languages. Legacy approaches recognize strings of phonemes as syllables, used by dictionary queries to see if a word can be found to uniquely identify a language. This dissertation research considers an alternative means of determining language identification of speech data based solely on analysis of frequency-domain data. An analytical approach to speech language identification by three comparative techniques is performed. First, a character-based pattern analysis is performed using the Rix and Forster algorithm to replicate their research on language identification. Second, techniques of phoneme recognition and their relative pattern of occurrence in speech samples are measured for performance in ability for language identification using the Rix and Forster approach. Finally, an experiment using statistical analysis of time-ensemble frequency spectrum data is assessed for its ability to establish spectral patterns for language identification, along with performance. This novel approach is applied to spectrogram audio data using pattern analysis techniques for language identification. It applies the Rix and Forster method to the ensemble of spectral frequencies used over the duration of a speech waveform. This novel approach is compared to the applications of the Rix and Forster algorithm to character-based and phoneme symbols for language identification on the basis of statistical accuracy, processing time requirements, and spatial processing resource needs. The audio spectrum analysis also demonstrates the ability to perform speaker identification using the same techniques performed for language identification. The results of this research demonstrate the efficacy of audio frequency-domain pattern analysis applied to speech waveform data. It provides an efficient technique in language identification without reliance upon linguistic approaches using phonemes or word derivations. This work also demonstrates a quick, automated means by which information gatherers, travelers, and diplomatic officials might obtain rapid language identification supporting time-critical determination of appropriate translator resource needs.

Spectrogram analysis of the most popular works of Adam Young’s four major projects; Owl City, Sky Sailing, Port Blue, and The Score Project. Spectrograms reveal several elements separate to what waveforms can show, and better show elements like frequency saturation, frequency ranges, overtones, and timbral sections. All these elements also can be used to better describe the phenomenon of the Autonomous Sensory Meridian Response (ASMR) and can also be seen on a spectrogram. Because of the ability to see these different elements, a spectrogram provides a good vehicle to analyze and compare elements of Young’s works and ASMR.After analysis, Young’s works show similar types of spectrograms to ASMR content, but the link between the influences between the two is not certain, it is probable that popular music and the ASMR phenomenon are linked in some measurable way. This thesis provides insight into how to further and continue music and ASMR research.

The ultimate goal of automatic speech recognition (ASR) research is to allow a computer to recognize speech in real-time, with full accuracy, independent of vocabulary size, noise, speaker characteristics or accent. Today, systems are trained to learn an individual speaker's voice and larger vocabularies statistically, but accuracy is not ideal. A small gap between actual speech and acoustic speech representation in the statistical mapping causes a failure to produce a match of the acoustic speech signals by Hidden Markov Model (HMM) methods and consequently leads to classification errors. Certainly, these errors in the low level recognition stage of ASR produce unavoidable errors at the higher levels. Therefore, it seems that ASR additional research ideas to be incorporated within current speech recognition systems. This study seeks new perspective on speech recognition. It incorporates a new approach for speech recognition, supporting it with wider previous research, validating it with a lexicon of 533 words and integrating it with a current speech recognition method to overcome the existing limitations. The study focusses on applying image processing to speech spectrogram images (SSI). We, thus develop a new writing system, which we call the Speech-Image Recogniser Code (SIR-CODE). The SIR-CODE refers to the transposition of the speech signal to an artificial domain (the SSI) that allows the classification of the speech signal into segments. The SIR-CODE allows the matching of all speech features (formants, power spectrum, duration, cues of articulation places, etc.) in one process. This was made possible by adding a Realization Layer (RL) on top of the traditional speech recognition layer (based on HMM) to check all sequential phones of a word in single step matching process. The study shows that the method gives better recognition results than HMMs alone, leading to accurate and reliable ASR in noisy environments. Therefore, the addition of the RL for SSI matching is a highly promising solution to compensate for the failure of HMMs in low level recognition. In addition, the same concept of employing SSIs can be used for whole sentences to reduce classification errors in HMM based high level recognition. The SIR-CODE bridges the gap between theory and practice of phoneme recognition by matching the SSI patterns at the word level. Thus, it can be adapted for dynamic time warping on the SIR-CODE segments, which can help to achieve ASR, based on SSI matching alone.

The Santa Barbara Instrument Group self-guided spectrograph is aimed primarily at the advanced amateur astronomer. This study has determined that this instrument is capable of being used for scientific research. A comparison lamp box was built to be used in conjunction with the spectrograph. A test was conducted to determine the accuracy of the grating positioning micrometer. A chart was created to allow the conversion of the micrometer readings to correlate with the position of the low-dispersion grating. To determine the scientific applicability of the spectrograph, it was used to: (1) measure the expansion rate of Nova Sagittarius 2004, (2) look for a 13.9-minute periodicity in the width and position of the H-alpha line of the binary star system X Persei, and (3) to measure the Doppler shift of X Persei. The instrument proved to be capable of being used to determine radial velocities and Doppler shifts.
Department of Physics and Astronomy

Nos travaux de recherches présentés dans ce manuscrit ont pour objectif, l'optimisation des performances des algorithmes de suivi des formants. Pour ce faire, nous avons commencé par l'analyse des différentes techniques existantes utilisées dans le suivi automatique des formants. Cette analyse nous a permis de constater que l'estimation automatique des formants reste délicate malgré l'emploi de diverses techniques complexes. Vue la non disponibilité des bases de données de référence en langue arabe, nous avons élaboré un corpus phonétiquement équilibré en langue arabe tout en élaborant un étiquetage manuel phonétique et formantique. Ensuite, nous avons présenté nos deux nouvelles approches de suivi de formants dont la première est basée sur l'estimation des crêtes de Fourier (maxima de spectrogramme) ou des crêtes d'ondelettes (maxima de scalogramme) en utilisant comme contrainte de suivi le calcul de centre de gravité de la combinaison des fréquences candidates pour chaque formant, tandis que la deuxième approche de suivi est basée sur la programmation dynamique combinée avec le filtrage de Kalman. Finalement, nous avons fait une étude exploratrice en utilisant notre corpus étiqueté manuellement comme référence pour évaluer quantitativement nos deux nouvelles approches par rapport à d'autres méthodes automatiques de suivi de formants. Nous avons testé la première approche par détection des crêtes ondelette, utilisant le calcul de centre de gravité, sur des signaux synthétiques ensuite sur des signaux réels de notre corpus étiqueté en testant trois types d'ondelettes complexes (CMOR, SHAN et FBSP). Suite à ces différents tests, il apparaît que le suivi de formants et la résolution des scalogrammes donnés par les ondelettes CMOR et FBSP sont meilleurs qu'avec l'ondelette SHAN. Afin d'évaluer quantitativement nos deux approches, nous avons calculé la différence moyenne absolue et l'écart type normalisée. Nous avons fait plusieurs tests avec différents locuteurs (masculins et féminins) sur les différentes voyelles longues et courtes et la parole continue en prenant les signaux étiquetés issus de la base élaborée comme référence. Les résultats de suivi ont été ensuite comparés à ceux de la méthode par crêtes de Fourier en utilisant le calcul de centre de gravité, de l'analyse LPC combinée à des bancs de filtres de Mustafa Kamran et de l'analyse LPC dans le logiciel Praat. D'après les résultats obtenus sur les voyelles /a/ et /A/, nous avons constaté que le suivi fait par la méthode ondelette avec CMOR est globalement meilleur que celui des autres méthodes Praat et Fourier. Cette méthode donne donc un suivi de formants (F1, F2 et F3) pertinent et plus proche de suivi référence. Les résultats des méthodes Fourier et ondelette sont très proches dans certains cas puisque toutes les deux présentent moins d'erreurs que la méthode Praat pour les cinq locuteurs masculins ce qui n'est pas le cas pour les autres voyelles où il y a des erreurs qui se présentent parfois sur F2 et parfois sur F3. D'après les résultats obtenus sur la parole continue, nous avons constaté que dans le cas des locuteurs masculins, les résultats des deux nouvelles approches sont notamment meilleurs que ceux de la méthode LPC de Mustafa Kamran et ceux de Praat même si elles présentent souvent quelques erreurs sur F3. Elles sont aussi très proches de la méthode par détection de crêtes de Fourier utilisant le calcul de centre de gravité. Les résultats obtenus dans le cas des locutrices féminins confirment la tendance observée sur les locuteurs
Our research work presented in this thesis aims the optimization of the performance of formant tracking algorithms. We began by analyzing different existing techniques used in the automatic formant tracking. This analysis showed that the automatic formant estimation remains difficult despite the use of complex techniques. For the non-availability of database as reference in Arabic, we have developed a phonetically balanced corpus in Arabic while developing a manual phonetic and formant tracking labeling. Then we presented our two new automatic formant tracking approaches which are based on the estimation of Fourier ridges (local maxima of spectrogram) or wavelet ridges (local maxima of scalogram) using as a tracking constraint the calculation of center of gravity of a set of candidate frequencies for each formant, while the second tracking approach is based on dynamic programming combined with Kalman filtering. Finally, we made an exploratory study using manually labeled corpus as a reference to quantify our two new approaches compared to other automatic formant tracking methods. We tested the first approach based on wavelet ridges detection, using the calculation of the center of gravity on synthetic signals and then on real signals issued from our database by testing three types of complex wavelets (CMOR, SHAN and FBSP). Following these tests, it appears that formant tracking and scalogram resolution given by CMOR and FBSP wavelets are better than the SHAN wavelet. To quantitatively evaluate our two approaches, we calculated the absolute difference average and standard deviation. We made several tests with different speakers (male and female) on various long and short vowels and continuous speech signals issued from our database using it as a reference. The formant tracking results are compared to those of Fourier ridges method calculating the center of gravity, LPC analysis combined with filter banks method of Kamran.M and LPC analysis integrated in Praat software. According to the results of the vowels / a / and / A /, we found that formant tracking by the method with wavelet CMOR is generally better than other methods. Therefore, this method provides a correct formant tracking (F1, F2 and F3) and closer to the reference. The results of Fourier and wavelet methods are very similar in some cases since both have fewer errors than the method Praat. These results are proven for the five male speakers which is not the case for the other vowels where there are some errors which are present sometimes in F2 and sometimes in F3. According to the results obtained on continuous speech, we found that in the case of male speakers, the result of both approaches are particularly better than those of Kamran.M method and those of Praat even if they are often few errors in F3. They are also very close to the Fourier ridges method using the calculation of center of gravity. The results obtained in the case of female speakers confirm the trend observed over the male speakers

Convolutional neural networks have been successfully applied to both object classification and audio classification. The aim of this thesis is to evaluate the degree of how well transfer learning of convolutional neural networks, trained in the object classification domain on large datasets (such as CIFAR-10, and ImageNet), can be applied to the audio classification domain when only a small dataset is available. In this work, four different convolutional neural networks are tested with three configurations of transfer learning against a configuration without transfer learning. This allows for testing how transfer learning and the architectural complexity of the networks affects the performance. Two of the models developed by Google (Inception-V3, Inception-ResNet-V2), are used. These models are implemented using the Keras API where they are pre-trained on the ImageNet dataset. This paper also introduces two new architectures which are developed by the authors of this thesis. These are Mini-Inception, and Mini-Inception-ResNet, and are inspired by Inception-V3 and Inception-ResNet-V2, but with a significantly lower complexity. The audio classification dataset consists of audio from RC-boats which are transformed into mel-spectrogram images. For transfer learning to be possible, Mini-Inception, and Mini-Inception-ResNet are pre-trained on the dataset CIFAR-10. The results show that transfer learning is not able to increase the performance. However, transfer learning does in some cases enable models to obtain higher performance in the earlier stages of training.