In this thesis, we consider the analysis of music and environmental audio recordings with neural networks. Recently, neural networks have been shown to be an effective family of models for speech recognition, computer vision, natural language processing and a number of other statistical modelling problems. The composite layer-wise structure of neural networks allows for flexible model design, where prior knowledge about the domain of application can be used to inform the design and architecture of the neural network models. Additionally, it has been shown that when trained on sufficient quantities of data, neural networks can be directly applied to low-level features to learn mappings to high level concepts like phonemes in speech and object classes in computer vision. In this thesis we investigate whether neural network models can be usefully applied to processing music and environmental audio. With regards to music signal analysis, we investigate 2 different problems. The fi rst problem, automatic music transcription, aims to identify the score or the sequence of musical notes that comprise an audio recording. We also consider the problem of automatic chord transcription, where the aim is to identify the sequence of chords in a given audio recording. For both problems, we design neural network acoustic models which are applied to low-level time-frequency features in order to detect the presence of notes or chords. Our results demonstrate that the neural network acoustic models perform similarly to state-of-the-art acoustic models, without the need for any feature engineering. The networks are able to learn complex transformations from time-frequency features to the desired outputs, given sufficient amounts of training data. Additionally, we use recurrent neural networks to model the temporal structure of sequences of notes or chords, similar to language modelling in speech. Our results demonstrate that the combination of the acoustic and language model predictions yields improved performance over the acoustic models alone. We also observe that convolutional neural networks yield better performance compared to other neural network architectures for acoustic modelling. For the analysis of environmental audio recordings, we consider the problem of acoustic event detection. Acoustic event detection has a similar structure to automatic music and chord transcription, where the system is required to output the correct sequence of semantic labels along with onset and offset times. We compare the performance of neural network architectures against Gaussian mixture models and support vector machines. In order to account for the fact that such systems are typically deployed on embedded devices, we compare performance as a function of the computational cost of each model. We evaluate the models on 2 large datasets of real-world recordings of baby cries and smoke alarms. Our results demonstrate that the neural networks clearly outperform the other models and they are able to do so without incurring a heavy computation cost.

Music signal analysis is a research field concerning the extraction of meaningful information from musical audio signals. This thesis analyses the music signals from the note-level to the song-level in a bottom-up manner and situates the research in two Music information retrieval (MIR) problems: audio onset detection (AOD) and music structural segmentation (MSS). Most MIR tools are developed for and evaluated on Western music with specific musical knowledge encoded. This thesis approaches the investigated tasks from a cross-cultural perspective by developing audio features and algorithms applicable for both Western and non-Western genres. Two Chinese Jingju databases are collected to facilitate respectively the AOD and MSS tasks investigated. New features and algorithms for AOD are presented relying on fusion techniques. We show that fusion can significantly improve the performance of the constituent baseline AOD algorithms. A large-scale parameter analysis is carried out to identify the relations between system configurations and the musical properties of different music types. Novel audio features are developed to summarise music timbre, harmony and rhythm for its structural description. The new features serve as effective alternatives to commonly used ones, showing comparable performance on existing datasets, and surpass them on the Jingju dataset. A new segmentation algorithm is presented which effectively captures the structural characteristics of Jingju. By evaluating the presented audio features and different segmentation algorithms incorporating different structural principles for the investigated music types, this thesis also identifies the underlying relations between audio features, segmentation methods and music genres in the scenario of music structural analysis.

Nowadays, the Electrical Bioimpedance (EBI) measurements have become a commonpractice as they are useful for different clinical applications. EBI technology and EBImeasurement systems are relatively simple when compared to other type of medicalinstrumentation. But even in such simple measurement systems measurement artifact mayoccur. One of the most common artifacts present measurements is the Hook Effect, which isidentifiable by the hook-alike deviation on the EBI data that it produces on the impedanceplot.The Hook Effect influences typical EBI data analysis processes like Cole fitting processand the estimation of the Cole parameters, which are critical for several EBI applications.Therefore the Hook Effect must be corrected, compensated or removed before the any fittingprocess.With the goal of finding a reliable correction method the origin and the impact on theEBI measurement of the Hook Effect is studied in this thesis. The currently used Tdcompensation method is also studied and a new approach for compensation and correction ispresented.The results indicate that the proposed method truly corrects the Hook Effect and that themethodology to select the correcting parameters is solid based on the origin of the Hook Effectand it is extracted from the EBI measurement it-self avoiding any external dependency.

The concept of Electrical Bio Impedance prevails in this thesis. The EBI measurement which is used for obtaining the body composition is, by virtue of time becoming of great use as its one of the easiest method of finding out the body composition. In simple words, EBI is the opposition offered by the body to the current. It is just like another analysis tool. The result is only as good as the test is done. In this thesis, we have done the analysis on the neonatal EBI measurements of two kinds.In this work, 293 measurements are obtained from 12 babies and 230 measurements are obtained from 7 babies have been analyzed with the purpose of obtaining reference values for the spectrum of complex EBI. The analysis uses both statistical and model approach of obtaining reference values and in order to fit the given data, Cole model analysis is used.Filters were applied to get the highest degree of correctness. In the due course of the filtering, it was found that the measurements from some babies have been deleted. The Standard Error of Estimation (S.E.E.) is a parameter used for obtaining the further reliable and most probable output. The further analysis is done using MATLAB and the results are been compared to the previous analysis report.

Electrical Bioimpedance (EBI) is the opposition offered by the biological material to theflow of electric current. Nowadays EBI technology is widely used for total body compositionand body fluid distribution.In this project a software suite is developed by using the GUI tool of Matlab, thissoftware is meant to help to remove artefacts from the EBI measurement and to visualize theEBIS measurements and the processing performed on them.Hook effect is one of the major artefacts found in EBI measurements, which createsproblems in any analysis. To eliminate the Hook effect some methods are followed. The data’sare processed using these methods and they are visualized. For the better understanding, bothraw data and the corrected data are plotted in impedance and admittance plots. The correcteddata is stored for further use and analysis.

No description available.

O controle de perdas em sistemas de distribuição de água para abastecimento é uma preocupação constante, e uma tarefa fundamental para a solução do problema é a detecção rápida e confiável dos vazamentos que frequentemente iniciam em qualquer ponto da rede. Uma abordagem promissória é a detecção de vazamentos baseada na análise de sinais adquiridos pelo monitoramento das redes durante sua operação, e dentro dela se enquadra este trabalho. É desenvolvido um sistema de reconhecimento de padrões para análise de sinais de pressão e vazão que permite identificar se durante a aquisição do sinal aconteceu um vazamento ou não. Para a conformação desse sistema diversas técnicas são exploradas, incluindo a extração de características no domínio do tempo (energia, entropia, número de cruzamentos por zero) e na decomposição wavelet (distribuição da energia nas componentes). Também é explorado o uso de algoritmos para classificação de diferentes tipos (vizinhos mais próximos, árvore de decisão, regra de decisão, Naive Bayes, máquina de vetor suporte e rede neural artificial com funções de base radial). Sinais são adquiridos junto ao circuito hidráulico experimental, que permitiu a simulação da ocorrência de um vazamento na rede, para constituir uma amplia base de dados com sinais de exemplo. Além da revisão bibliográfica e os conceitos relativos às metodologias exploradas, são apresentadas neste documento as análises que conduzem à criação do sistema de reconhecimento de padrões mais apropriado para o problema. Das análises dos diferentes métodos considerados é definido o sistema de reconhecimento de padrões, em suas etapas de segmentação e padronização, extração de características e classificação. A avaliação do sistema proposto mostra um desempenho totalmente satisfatório, com reconhecimento acertado de sinais vinculados ou não a um vazamento em mais de 95% dos testes. / Control of losses in water supply systems is a constant concern, and a key to the solution of this problem is the rapid and reliable detection of leaks that often begin anywhere on the network. A promising approach to solve the problem is the leak detection based on the analysis of signals acquired by monitoring the network in operation, and this research fits with that approach. Its developed a pattern recognition system for the analysis of pressure and flow signals, which identifies whether a leak happened during signal acquisition. Several techniques are exploited for forming this system, including the feature extraction in the time domain (energy, entropy, zero crossings count) and in the wavelet decomposition (energy distribution in the components). Also, the use of different types of algorithms for classification (nearest neighbors, decision tree, decision rule, Naive Bayes, support vector machine and artificial neural network with radial basis functions) is explored. Signals are acquired from the experimental hydraulic circuit, allowing the simulation of the onset of a leak in the network, to form a big database of example signals. Besides the literature review and the concepts relating to the considered methods, in this document are shown the analyses leading to the creation of the pattern recognition system most appropriate for the problem. The analysis of the methods considered allows defining the pattern recognition system, which is composed by segmentation and standardization, feature extraction and classification. The evaluation of the proposed system shows a completely satisfactory performance, recognizing rightly the signals as linked or not to a leak in more than 95% of the tests.

<p>The thesis presents an investigation of the potential of measuring plant condition from hyperspectral reflectance data. To do this, some linear methods for embedding the high dimensional hyperspectral data and to perform regression to a plant condition space have been compared. A preprocessing step that aims at normalized illumination intensity in the hyperspectral images has been conducted and some different methods for this purpose have also been compared.A large scale experiment has been conducted where tobacco plants have been grown and treated differently with respect to watering and nutrition. The treatment of the plants has served as ground truth for the plant condition. Four sets of plants have been grown one week apart and the plants have been measured at different ages up to the age of about five weeks. The thesis concludes that there is a relationship between plant treatment and their leaves' spectral reflectance, but the treatment has to be somewhat extreme for enabling a useful treatment approximation from the spectrum. CCA has been the proposed method for calculation of the hyperspectral basis that is used to embed the hyperspectral data to the plant condition (treatment) space. A preprocessing method that uses a weighted normalization of the spectrums for illumination intensity normalization is concluded to be the most powerful of the compared methods.</p>

The thesis presents an investigation of the potential of measuring plant condition from hyperspectral reflectance data. To do this, some linear methods for embedding the high dimensional hyperspectral data and to perform regression to a plant condition space have been compared. A preprocessing step that aims at normalized illumination intensity in the hyperspectral images has been conducted and some different methods for this purpose have also been compared.A large scale experiment has been conducted where tobacco plants have been grown and treated differently with respect to watering and nutrition. The treatment of the plants has served as ground truth for the plant condition. Four sets of plants have been grown one week apart and the plants have been measured at different ages up to the age of about five weeks. The thesis concludes that there is a relationship between plant treatment and their leaves' spectral reflectance, but the treatment has to be somewhat extreme for enabling a useful treatment approximation from the spectrum. CCA has been the proposed method for calculation of the hyperspectral basis that is used to embed the hyperspectral data to the plant condition (treatment) space. A preprocessing method that uses a weighted normalization of the spectrums for illumination intensity normalization is concluded to be the most powerful of the compared methods.

Parameter identification of an induction machine is of great importance in numerous industrial applications, including the assessment of machine performance and design of control schemes. Parameter identification is based on the input-output signals and the model used. Many researches have applied the inverter drive to control the exciting signal of the induction machine in the identifying process. This study proposed a method to identify all parameter of the induction machine with a no-load low-voltage starting test. The method has a simple structure without needing extra hardware, which could significantly simplify the procedures and save cost. Based on the curves of resistance and reactance, the user can obtain the machine¡¦s equivalent circuit parameters. With the identified parameters of the equivalent circuit, input voltage, and rotor speed, the user can find the torque. From the torque and rotor speed, the user can find the mechanical parameters. A least mean square (LMS) method was used with a particle swarm optimization (PSO) method to solve the aforementioned problem. From various tests, the practicability and accuracy of this method can been proven. This study also proposes a method to rapidly analyze power parameters. This method uses two adjacent data to compute the fundamental frequency component of voltage or current. The parameters of fundamental frequency component include frequency, amplitude, and phase. Under the condition of varied parameters, the frequency and phase are dependent. This method fixes the frequency and computes the amplitude and phase, and then stable results will be obtained.