With the development of big data computing, the speech recognition has been popular for serving human’s life. However, when place the speech recognition system into noisy environments, the background noises greatly degrades the speech recognition system accuracy as it adds in unuseful information into the desired speech. Thus for a speech recognition system, obtaining a good performance under noises has become a vital issue. To tackle the noise effect problem of automatic speech recognition (ASR), a method to reduce the noise effect is essential. Recently, multiple of methods have been developed to enhance the speech signal, they usually follow the principle of suppressing the noise in a noisy speech signal. This thesis simulated the popular techniques for speech recogniton and speech enhancement, which are the multilayer perceptron and the spectral subtraction. The aim of this work is to use MATLAB to build an automatic speech recognition system that can be used in noisy environment. MATLAB simulations are used to verify the success of recognition with clean speech and show the system performance improvements after applying speech enhancement method in seven kinds of noisy environments. The result is presented by using comparative histograms between noisy signals and corresponding denoised signals. It shows that, using denoised signal will obtain a higher recognition rate, thus we can say the system performance is improved in noisy environments.

No description available.

Human fingerprints are rich in details called the minutiae, which can be used as identification marks for fingerprint verification. To get the details, the fingerprint capturing techniques are to be improved. Since when we the fingerprint is captured, the noise from outside adds to it. The goal of this thesis is to remove the noise present in the fingerprint image. To achieve a good quality fingerprint image, this noise has to be removed or suppressed and here it is done by using an algorithm or technique called ’Spectral Subtraction’, where the algorithm is based on subtraction of estimated noise spectrum from noisy signal spectrum. The performance of the algorithm is assessed by comparing the original fingerprint image and image obtained after spectral subtraction several parameters like PSNR, SSIM and also for different fingerprints on the database. Finally, performance matching was done using NIST matching software, and the obtained results were presented in the form of Receiver Operating Characteristics (ROC)graphs, using MATLAB, and the experimental results were presented.

Este trabalho tem a finalidade de apresentar e comparar técnicas de redução de ruído utilizando como critérios de avaliação a mínima distorção espectral e a redução de ruído, na reconstrução dos sinais de voz degradados por ruído. Para tanto, utilizou-se os filtros de Kalman de tempo discreto e de freqüência discreta em conjunto com a técnica de subtração espectral de potência. Os sinais utilizados foram contaminados por ruídos branco e colorido, e a avaliação do desempenho dos algoritmos foi realizada tendo-se como parâmetros a relação sinal/ruído segmentada (SNRseg) e a distância de Itakura-Saito (d(a,b)). Após o processamento, verificou-se que a técnica, proposta neste trabalho, de filtragem de Kalman no tempo em conjunto com a subtração espectral de potência, apresentou resultados um pouco melhores em relação à filtragem de Kalman na freqüência em conjunto com a subtração espectral de potência. / This work has as main objective to present and to compare techniques of noise reduction using as evaluation criterion the low spectral distortion and the noise reduction in the reconstruction of corrupted speech signals. For so much, it was used the Kalman\'s filters in the time and frequency domain together with the technique of power spectral subtraction. The used signals were corrupted by white and colored noises and the evaluation of effectiveness of the algorithms was accomplished using the segmental signal-to-noise ratio (SNRseg) and the Itakura-Saito distance (d(a,b)). After the processing, it was noticed that the Kalman filtering in the time together with power spectral subtraction presented better results than the Kalman filtering in the frequency together with power spectral subtraction.

"Speech enhancement in an automobile is a challenging problem because interference can come from engine noise, fans, music, wind, road noise, reverberation, echo, and passengers engaging in other conversations. Hands-free microphones make the situation worse because the strength of the desired speech signal reduces with increased distance between the microphone and talker. Automobile safety is improved when the driver can use a hands-free interface to phones and other devices instead of taking his eyes off the road. The demand for high quality hands-free communication in the automobile requires the introduction of more powerful algorithms. This thesis shows that a unique combination of five algorithms can achieve superior speech enhancement for a hands-free system when compared to beamforming or spectral subtraction alone. Several different designs were analyzed and tested before converging on the configuration that achieved the best results. Beamforming, voice activity detection, spectral subtraction, perceptual nonlinear weighting, and talker isolation via pitch tracking all work together in a complementary iterative manner to create a speech enhancement system capable of significantly enhancing real world speech signals. The following conclusions are supported by the simulation results using data recorded in a car and are in strong agreement with theory. Adaptive beamforming, like the Generalized Side-lobe Canceller (GSC), can be effectively used if the filters only adapt during silent data frames because too much of the desired speech is cancelled otherwise. Spectral subtraction removes stationary noise while perceptual weighting prevents the introduction of offensive audible noise artifacts. Talker isolation via pitch tracking can perform better when used after beamforming and spectral subtraction because of the higher accuracy obtained after initial noise removal. Iterating the algorithm once increases the accuracy of the Voice Activity Detection (VAD), which improves the overall performance of the algorithm. Placing the microphone(s) on the ceiling above the head and slightly forward of the desired talker appears to be the best location in an automobile based on the experiments performed in this thesis. Objective speech quality measures show that the algorithm removes a majority of the stationary noise in a hands-free environment of an automobile with relatively minimal speech distortion."

Este trabalho tem a finalidade de apresentar e comparar técnicas de redução de ruído utilizando como critérios de avaliação a mínima distorção espectral e a redução de ruído, na reconstrução dos sinais de voz degradados por ruído. Para tanto, utilizou-se os filtros de Kalman de tempo discreto e de freqüência discreta em conjunto com a técnica de subtração espectral de potência. Os sinais utilizados foram contaminados por ruídos branco e colorido, e a avaliação do desempenho dos algoritmos foi realizada tendo-se como parâmetros a relação sinal/ruído segmentada (SNRseg) e a distância de Itakura-Saito (d(a,b)). Após o processamento, verificou-se que a técnica, proposta neste trabalho, de filtragem de Kalman no tempo em conjunto com a subtração espectral de potência, apresentou resultados um pouco melhores em relação à filtragem de Kalman na freqüência em conjunto com a subtração espectral de potência. / This work has as main objective to present and to compare techniques of noise reduction using as evaluation criterion the low spectral distortion and the noise reduction in the reconstruction of corrupted speech signals. For so much, it was used the Kalman\'s filters in the time and frequency domain together with the technique of power spectral subtraction. The used signals were corrupted by white and colored noises and the evaluation of effectiveness of the algorithms was accomplished using the segmental signal-to-noise ratio (SNRseg) and the Itakura-Saito distance (d(a,b)). After the processing, it was noticed that the Kalman filtering in the time together with power spectral subtraction presented better results than the Kalman filtering in the frequency together with power spectral subtraction.

The objective of this thesis is to build a phrase recognition system under noisy environment that can be used in real-life. In this system, the noisy speech is first filtered by the enhanced spectral subtraction method to reduce the noise level. Then the MFCC with cepstral mean subtraction is applied to extract the speech features. Finally, hidden Markov model (HMM) is used in the last stage to build the probabilistic model for each phrase. A Mandarin microphone database of 514 company names that are in Taiwan¡¦s stock market is collected. A speaker independent noisy phrase recognition system is then implemented. This system has been tested under various noise environments and different noise strengths.

La prothèse cochléaire est un appareillage destiné à la réhabilitation des surdités profondes et totales dont un appareillage conventionnel est inefficace. Elle assure la stimulation directe des neurones cochléaires à travers un faisceau d’électrodes. Différents travaux de recherches ont été établis afin d'évaluer l'intelligibilité de la parole chez les sujets bilatéralement implantés en environnements silencieux et bruité. Les résultats ont montré une bonne intelligibilité de la parole en milieu silencieux. Toutefois, les capacités de perception de la parole chez les patients implantés se dégradent en environnement bruité. Nous avons de ce fait proposé trois approches de traitement du signal en vue d'une amélioration de l'intelligibilité de la parole dans le cas de l'implant cochléaire bilatéral: la stimulation bilatérale décalée, l'algorithme de la soustraction spectrale bi-voie et l'algorithme de la soustraction interspectrale. Des améliorations de l'intelligibilité de la parole entre 4% et 10% ont été notées dans le cas de la stimulation bilatérale décalée par rapport à la stimulation bilatérale symétrique. L'approche basée sur l'algorithme de la soustraction spectrale bi-voie présentait des améliorations variables entre 10% et 17%. De meilleures performances ont été obtenues lorsque l'approche basée sur l'algorithme de la soustraction interspectrale est considérée où les améliorations étaient entre 15% et 27% / Cochlear prostheses are intended for persons suffering from deep or total deafness where conventional prostheses proved ineffective. In quiet listening conditions, most bilateral cochlear implant (BCI) users can now achieve even more than 80% word recognition scores regardless the used device. However, under more challenging listening conditions, BCI recipients perform poorly, compared to normal-hearing listeners. In this work, we proposed three speech processing approaches for speech intelligibility improvement. The first is based on shifted bilateral cochlear implant stimulation; the second is based on dual-channel spectral subtraction algorithm and finally the cross power spectral subtraction algorithm was considered. Experimental results showed a speech intelligibility improvement between 4% and 10% when the shifted bilateral cochlear implant stimulation is considered. Performance amelioration was observed when the dual-channel spectral subtraction based speech enhancement algorithm was considered and the improvement was between 10% and 17%. The better performance was obtained when noisy speech signals were processed using cross power spectral subtraction algorithm and the improvement was between 15% and 27%

More than one million fetal deaths occur in the United States every year [1]. Monitoring the long-term heart rate variability provides a great amount of information about the fetal health condition which requires continuous monitoring of the fetal heart rate. All the existing technologies have either complex instrumentation or need a trained professional at all times or both. The existing technologies are proven to be impractical for continuous monitoring [2]. Hence, there is an increased interest towards noninvasive, continuous monitoring, and less expensive technologies like fetal phonocardiography. Fetal Phonocardiography (FPCG) signal is obtained by placing an acoustic transducer on the abdomen of the mother. FPCG is rich in physiological bio-signals and can continuously monitor the fetal heart rate non-invasively. Despite its high diagnostic potential, it is still not being used as the secondary point of care. There are two challenges as to why it is still being considered as the secondary point of care; in the data acquisition system and the signal processing methodologies. The challenges pertaining to data acquisition systems are but not limited to sensor placement, maternal obesity and multiple heart rates. While, the challenges in the signal processing methodologies are dynamic nature of FPCG signal, multiple known and unknown signal components and SNR of the signal. Hence, to improve the FPCG based care, challenges in FPCG signal processing methodologies have been addressed in this study. A comparative evaluation was presented on various advanced signal processing techniques to extract the bio-signals with fidelity. Advanced signal processing approaches, namely empirical mode decomposition, spectral subtraction, wavelet decomposition and adaptive filtering were used to extract the vital bio-signals. However, extracting these bio-signals with fidelity is a challenging task in the context of FPCG as all the bio signals and the unwanted artifacts overlap in both time and frequency. Additionally, the signal is corrupted by noise induced from the fetal and maternal movements as well the background and the sensor. Empirical mode decomposition algorithm was efficient to denoise and extract the maternal and fetal heart sounds in a single step. Whereas, spectral subtraction was used to denoise the signal which was later subjected to wavelet decomposition to extract the signal of interest. On the other hand, adaptive filtering was used to estimate the fetal heart sound from a noisy FPCG where maternal heart sound was the reference input. The extracted signals were validated by obtaining the frequency ranges computed by the Short Time Fourier Transform (STFT). It was observed that the bandwidths of extracted fetal heart sounds and maternal heart sounds were consistent with the existing gold standards. Furthermore, as a means of additional validation, the heart rates were calculated. Finally, the results obtained from all these methods were compared and contrasted qualitatively and quantitatively.

Les implants cochléaires sont des appareils destinés à la réhabilitationdes surdités profondes et totales. Ils assurent la stimulation du nerf auditif en plaçant des électrodes dans la cochlée. Différentes études ont été établis afin d’améliorer l’intelligibilité de la parole dans le bruit chez le patientporteur de cet appareil. Les techniques bilatérales et binaurales permettent dereproduire une audition binaurale, car les deux oreilles sont simulées (commepour les personnes normo-entendantes). Ainsi la localisation et la perceptiondes sons environnants sont améliorées par rapport à une implantationmonaurale. Toutefois, les capacit´es de reconnaissances des mots sont trèsvite limitées en pr´esence de bruits. Nous avons d´evelopp´es des techniquesde r´eduction de bruit afin d’augmenter les performances de reconnaissance.Des améliorations de 10% à 15% suivant les conditions ont été observées. Néanmoins, si la perception est améliorée par les algorithmes, ils focalisent sur une direction, et ainsi, la localisation est alors réduite à l’angle delocalisation. Une seconde étude a alors été effectuée pour mesurer l’effetdes algorithmes sur la localisation. Ainsi, le beamformer donne les meilleurs résultats de compréhension mais la moins bonne localisation. La ré-injectiond’un pourcentage du signal d’entrée sur la sortie a permis de compenser laperte de la localisation sans détériorer l’intelligibilité. Le résultat de ces deux expériences montre qu’il faut un compromis entre laperception et la localisation des sons pour obtenir les meilleures performances. / Cochlear implants are to sail for the rehabilitation of deep and totaldeafness. They provide stimulation of the auditory nerve by placing electrodesin the cochlea. Various studies have been established to improve thespeech intelligibility in noise in the patient of this device. Bilateral andbinaural techniques allow reproducing a binaural hearing, since both earsare simulated (as for normal hearing people). Thus localization and theperception of the surrounding sounds are improved from a monauralimplantation. However, the recognition of the words capabilities are limitedvery quickly in the presence of noise.We developed noise reduction techniquesto increase the performance of recognition. Improvements of 10% to 15%depending on the conditions were observed. Nevertheless, if the perception isenhanced by algorithms, they focus on a direction, and thus the location isthen reduced at the corner of localization. Then, a second study was madeto measure the effect of localization algorithms. Thus, the beamformer givesthe best results of understanding but the less good location. The re-injectionof a percentage of the input to the output signal helped offset the loss of thelocation without damaging the intelligibility.The result of these two experiments shows that it takes a compromisebetween perception and sound localization for best performance.