In most speech communication systems, the presence of background noise causes the quality and intelligibility of speech to degrade, especially when the Signal-to-Noise Ratio (SNR) is low. Numerous speech enhancement techniques have been employed successfully in many applications. However, at low signal-to-noise ratios most of these speech enhancement techniques tend to introduce a perceptually annoying residual noise known as "musical noise". The research presented in this thesis aims to minimize this musical noise and maximize the noise reduction ability of speech enhancement algorithms to improve speech quality in low SNR environments. This thesis proposes two novel speech enhancement algorithms based on Weiner and Kalman filters, and exploit the masking properties of the human auditory system to reduce background noise. The perceptual Wiener filter method uses either temporal or simultaneous masking to adjust the Wiener gain in order to suppress noise below the masking thresholds. The second algorithm involves reshaping the corrupted signal according to the masking threshold in each critical band, followed by Kalman filtering. A comparison of the results from these proposed techniques with those obtained from traditional methods suggests that the proposed algorithms address the problem of noise reduction effectively while decreasing the level of the musical noise. In this thesis, many other existing competitive noise suppression methods have also been discussed and their performance evaluated under different types of noise environments. The performances were evaluated and compared to each other using both objective PESQ measures (ITU-T P.862) and subjective listening tests (ITU-T P.835). The proposed speech enhancement schemes based on the auditory masking model outperformed the other methods that were tested.

This thesis is directed at exploiting information in the coda of seismic phases and the ambient noise field to provide new constraints on the structure of the Australian Continent. ¶ The exploitation of the immediate coda following the onset of P waves from a distant earthquake using radial receiver functions is now a well established method. The 40 sec interval following P contains reverberations and conversions, by deconvolving the radial component trace with the vertical components, the conversions are emphasized by canceling the part of the response that are common to both components. A member of different styles of such deconvolution, are investigated and a variant of the multitaper method is adopted for subsequent applications. The TASMAL experiment 2003-2005 spans the expected location of the transition between Precambrian and Phanerozoic Australia. The 20 portable broadband stations were exploited in receiver function studies to extract S wave crustal structure through the inversion of stacked receiver functions using the Neighbourhood Algorithm. There is no clear crustal transition associated with the presence of Tasman Line. The Precambrian Cratons tend to exhibit crustal thicknesses close to 40 km but such values are also found in some Phanerozoic sites. ¶ The second part of the thesis is directed at the exploitation of ambient noise or seismic coda to gain information on the Green's function between seismic stations. The TASMAL experiment covered a significant fraction of the Australian continent with a simultaneous deployment of portable broadband stations. From these continuous records, it has proved possible to extract very clear Rayleigh wave signals for station separations up to 2000 km, and to demonstrate the frequency dependent variations in group velocity behaviour. The combination of the paths between the 20 stations localize such behaviour, but detailed images needed more data. The entire archive of portable broadband data recorded by RSES was mined, and combined with data from permanent stations to provide more than 1100 estimates of interstation Green's functions within Australia. Group velocity analysis as function of frequency was followed by nonlinear tomography with the Fast Marching Method. The resulting images of group velocity patterns as a function frequency show pronounced regions of lowered group velocities, most of which match regions of thick sediment. The frequency dependence is not consistent with just sedimentary structure and low midcrustal velocities, most likely due to elevated temperatures, are also needed. ¶ The surface wave portion of the interstation Green's function is the most energetic, and is normally all that seen in ambient noise studies. However, in the coda of events record at the broadband Warramunga seismic array in the Northern Territory, the P and S body wave components also emerge. The characteristics of these arrivals match those observed from nearby small earthquakes. The stacked cross-correlation is the normal approach to enhance Green's function information from ambient noise, but a broader spectral band width with the same phase response can be found by spectral division. It appears advantageous to compare both approaches and select the best result, since very little modifications to procedures are needed. ¶ The properties of the ambient noise at a single station have been investigated in the logarithmic spectral domain and a station dependent signal can be extracted by stacking. The signal appears to be related to the local structure beneath the station, and when fully characterized may provide a new means of investigating structure.

The need to attenuate noise transmitted into enclosed spaces such as aircraft cabins, automobiles and mining cabins has provided the impetus for many active noise control studies. Studies into active interior noise control began with a pressure squared cost function utilising multiple error sensors and control sources in an attempt to produce global control of the interior sound field. This work found problems with observability of the primary disturbances and a large number of error sensors and control sources were required to produce global control. Since this early work in the 1980's, many new acoustic based cost functions have been developed to improve on the performance of the pressure squared cost function. This thesis will focus on one novel acoustic cost function, virtual error sensing. Virtual error sensing is a relatively new technique which produces localised zones of attenuation at a location remote to the physical sensors. The practical advantage of this method is the people within these enclosed spaces are able to observe a reduction in sound pressure level without their movement being restricted by error sensors located close to their ears. The aim of this thesis is to further investigate the performance of forward-difference virtual error sensors in order to understand the factors that affect the accuracy of the pressure prediction at the virtual location and use this information to develop more accurate and efficient forward- difference virtual sensors. These virtual sensors use linear arrays of microphones containing two or more microphone elements and a linear or quadratic approximation is used to predict the sound at the virtual location. The prediction method determines the weights applied to each microphone signal to predict the sound pressure level at the virtual location. This study investigates susceptibility of the sensors to corruption as a result of phase and sensitivity mismatch between the microphones, as well as in the location of the elements in the error sensing array. A thorough error analysis of the forward-difference virtual microphones was performed in a one-dimensional sound field and in a plane wave sound field. The accuracy of the quadratic virtual microphone was found to be strongly affected by the presence of short wavelength extraneous noise. From this study, two novel virtual error sensing techniques were developed, namely; higher-order virtual sensors and adaptive virtual sensors. The higher-order virtual error sensors still employ the linear and quadratic prediction method but extra microphone elements are added to the array. The aim of these higher-order virtual microphones is to produce a more accurate prediction of the pressure at the virtual location by spatially filtering out any short wavelength extraneous noise that may corrupt the prediction. These virtual sensors were tested in a realtime control scenario in both a one-dimensional reactive sound field and in a free field. This work found that the higher-order virtual microphones can improve the prediction accuracy of the original virtual sensors but are still prone to problems of phase, sensitivity and position errors. Finally, the adaptive LMS virtual sensors were investigated in a SIMULINK simulation and tested experimentally using real-time control in a one-dimensional sound field. It was hoped that an adaptive LMS algorithm could overcome previous difficulties arising from inherent and transducer errors by adapting the weights of the signals from the sensing elements which form the array. The algorithm adapts the sensing microphone signals to produce the same signal as the microphone at the virtual location. Once this has been achieved, the sensing microphone weights are fixed and the microphone at the virtual location is removed, thus creating a virtual microphone. The SIMULINK simulation allowed the performance of the fixed weight and virtual microphones to be investigated in the presence of only phase errors, sensitivity errors and position errors and in the presence of all three combined. This work showed that the adaptive virtual sensors had the ability to compensate for the errors. The number of modes used in the simulations was varied to observe the performance of all virtual sensors in the presence of higher-order modes. The prediction accuracy of the fixed weight virtual sensors was found to be greatly affected by the presence of higher-order modes. The use of the adaptive virtual microphones to produce localised zones of quiet was examined experimentally using real-time control. The study found the real-time control performance is superior to that of the fixed weight higher-order virtual microphones and the original forward-difference virtual microphones. / Thesis (Ph.D.)--School of Mechanical Engineering, 2003.

Large electric transformers such as those used in high voltage substations radiate an annoying low frequency hum into nearby communities. Attempts have been made to actively control the noise by placing a large number of loudspeakers as control sources around noisy transformers to cancel the hum. These cancellation systems require a large number of loudspeakers to be successful due to the imposing size of the transformer structures. Thus such systems are very expensive if global noise reduction is to be achieved. The aim of this thesis is to investigate theoretically and experimentally the use of thin perforated panels closely placed to a heavy structure to reduce the radiation of unwanted harmonic noise. These panels can themselves be vibrated to form a control source radiating over a large surface surrounding the primary source. The problem of the equipment overheating inside the enclosure is alleviated because the holes in the panels still allow natural cooling. An initial study is carried out to determine the resonance frequencies of perforated panels. The use of previously determined effective elastic properties of the panels and Finite Element Analysis to theoretically calculate their resonance frequencies is examined. Secondly the attenuation provided by active noise control using perforated panels as control sources is explored by use of a coupled analysis, where the primary source is assumed to influence the radiation of the perforated control panel. This analysis was found to predict poorly the amount of attenuation that could be achieved, so an uncoupled analysis is undertaken, where both the primary and control sources are assumed to radiate independently of each other. Not only does this greatly simplify the theoretical analysis but it also enables prediction of attenuation levels which are comparable to those determined experimentally. The theoretical model is reformulated to enable comparison of the sound power attenuation provided by perforated panel control sources with that of traditional acoustic and structural control sources. Finally, the use of modal filtering of traditional acoustic error sensor signals to give transformed mode (or power mode) sensors is examined. The independently radiating acoustic transformed modes of the panel are determined by an eigenanalysis and a theoretical analysis is presented for a farfield acoustic power sensor system to provide a direct measurement of the total radiated acoustic power. The frequency dependence of the sensor system, and the amount of global sound power attenuation that can be achieved is examined. Experimental measurements are made to verify the theoretical model and show that a sound power sensor implemented with acoustic sensors can be used in a practical active noise control system to increase the amount of attenuation that can be achieved. Alternatively the sound power sensor can be used to reduce the number of error channels required by a control system to obtain a given level of attenuation when compared to traditional error criteria. The power mode sensor analysis is then applied to the perforated panel control system, with similar results. / Thesis (Ph.D.)--Engineering (Department of Mechanical Engineering), 1996.

Large electric transformers such as those used in high voltage substations radiate an annoying low frequency hum into nearby communities. Attempts have been made to actively control the noise by placing a large number of loudspeakers as control sources around noisy transformers to cancel the hum. These cancellation systems require a large number of loudspeakers to be successful due to the imposing size of the transformer structures. Thus such systems are very expensive if global noise reduction is to be achieved. The aim of this thesis is to investigate theoretically and experimentally the use of thin perforated panels closely placed to a heavy structure to reduce the radiation of unwanted harmonic noise. These panels can themselves be vibrated to form a control source radiating over a large surface surrounding the primary source. The problem of the equipment overheating inside the enclosure is alleviated because the holes in the panels still allow natural cooling. An initial study is carried out to determine the resonance frequencies of perforated panels. The use of previously determined effective elastic properties of the panels and Finite Element Analysis to theoretically calculate their resonance frequencies is examined. Secondly the attenuation provided by active noise control using perforated panels as control sources is explored by use of a coupled analysis, where the primary source is assumed to influence the radiation of the perforated control panel. This analysis was found to predict poorly the amount of attenuation that could be achieved, so an uncoupled analysis is undertaken, where both the primary and control sources are assumed to radiate independently of each other. Not only does this greatly simplify the theoretical analysis but it also enables prediction of attenuation levels which are comparable to those determined experimentally. The theoretical model is reformulated to enable comparison of the sound power attenuation provided by perforated panel control sources with that of traditional acoustic and structural control sources. Finally, the use of modal filtering of traditional acoustic error sensor signals to give transformed mode (or power mode) sensors is examined. The independently radiating acoustic transformed modes of the panel are determined by an eigenanalysis and a theoretical analysis is presented for a farfield acoustic power sensor system to provide a direct measurement of the total radiated acoustic power. The frequency dependence of the sensor system, and the amount of global sound power attenuation that can be achieved is examined. Experimental measurements are made to verify the theoretical model and show that a sound power sensor implemented with acoustic sensors can be used in a practical active noise control system to increase the amount of attenuation that can be achieved. Alternatively the sound power sensor can be used to reduce the number of error channels required by a control system to obtain a given level of attenuation when compared to traditional error criteria. The power mode sensor analysis is then applied to the perforated panel control system, with similar results. / Thesis (Ph.D.)--Engineering (Department of Mechanical Engineering), 1996.

Thesis (Ph. D.)--Ohio University, June, 1990. / Title from PDF t.p.

Thesis (M. Env. St.)--University of Adelaide, 1984. / Includes bibliographical references.

Thesis (Ph. D.)--University of New South Wales, 2005. / Also available online.

Diss. Linköping : Univ., 2003.

Thesis (Ph.D.)-- University of Adelaide, Dept. of Mechanical Engineering, 2000. / Bibliography: leaves 182-190. Also available electronically.