Differential microphone array for speech recognition

Speech recognition applications embedded on a PDA are already available on the market. The usual hardware for this kind of systems is a single microphone mounted on the PDA, giving good results within quiet environments. Though, the recognition rate falls drastically as the signal to noise ratio decreases. Arrays of microphones are then particularly interesting, allowing the discrimination of useful sounds sources within parasitic ones, thanks to an improved directivity pattern. In speech recognition applications on PDA, a trade-off is to be found on an array small enough to guarantee the ergonomics of the whole system, while operating on a bandwidth covering the frequency range of human voice, from 300 Hz up to 6 kHz and low distortion ratio. Differential arrays are well adapted to fulfill these specifications, since they are known to be robust and allowing small dimensions for a high directivity index. The performances could be improved by the addition of adaptive post-filtering and noise reduction algorithms. This work describes the design and the implementation of a recording device dedicated to speech recognition applications on PDA, based on overlapping differential arrays. The assessment of its performances in noisy environment are carried out and show the system efficiency.

Compensation des non-linéarités des systèmes haut-parleurs à pavillon

Delphine Bard

The assumptions of linearity and autonomy assumed in circuit theory are known to be respected only partially by electroacoustic transducers. In particular, the dynamics range of the latter is limited by the audible nonlinear effects that they are causing. In the domain of active noise control, the distortion products limit noise reduction performance. The present thesis work stems from characterization of these observations. We want to design a real-time system compensating the nonlinearities of electroacoustic devices. The first part of this report concerns the nonlinearity effects. It is shown that the use of classical characterization methods of an electroacoustic device (harmonic distortion, intermodulation, frequency difference ...) is limited. Hence, they can not be used to determine the nonlinearity laws themselves, but rather only their effects for arbitrary excitations. First, a method based on multitone harmonic excitations has been proposed and validated. It has been applied for the characterization of loudspeaker prototypes using several different technologies, which are used for active noise reduction in an aircraft turboreactor. The second part of this work addresses the elaboration and validation of a nonlinearity effect compensation method. This method is based on the description of the nonlinearities by the Volterra series. For this, we need to place an upstream system, characterized by the inverse nonlinearity law of the loudspeaker system. After having determined precisely the Volterra kernels of the system to be characterized, the kernels of the upstream compensation system are determined by assuming that their cascade obeys to a linear law. For this purpose, a kernel measurement method in the frequency domain has been developed, validated and tested. Since we are concerned with loudspeakers, we took their flying time into account. The compensation method has been validated by computing the simulated response of the compensation circuit. Resulting analog signals have then been applied to the loudspeaker. Measured performances fulfill the expectations. In the last part, the compensation method has been applied in a real-time Digital Signal Processing) DSP controller, which allowed to realize a demonstrator to the implemented for future industrial applications.

EPFL2005

Multiple scattering in wide-field optical coherence tomography

Boris Karamata

Optical Coherence Tomography (OCT), a well-established imaging method based on low-coherence interferometry, provides cross-sectional images of the internal structure of biological samples with a resolution in the micrometer range. OCT was successfully applied on various tissues such as for instance the retina, the skin or a tooth. In highly scattering tissues like the skin, probing depth is limited to approximately 2mm, mainly due to insufficient rejection of multiply scattered light. Presently, the contribution of multiple scattering in OCT is not fully understood. Therefore, there is a strong and urgent need to develop models allowing a reliable evaluation of the system's limitations as well as the improvement of the imaging capabilities. It is generally believed that a relevant model should account for loss of correlation between the reference and the sample field due to multiple scattering. We developed a new comprehensive model of OCT. Our preliminary study revealed that the reference and sample fields are actually fully correlated. This important result allowed us to model the OCT signal as a sum of stationary random phasors and treated it as a statistical signal. The mean of this signal can be calculated thanks to classical results of statistical optics and to a Monte Carlo simulation. Unlike other existing models, our model accounts for the source autocorrelation function. The model proved to be in excellent agreement with a whole range of experimental data gathered in a comprehensive study of cross-talk in wide-field OCT. Moreover, our results put in question the applicability of widely used models of OCT based on the "extended Huygens-Fresnel principle", which assume a partial correlation between interfering fields due to multiple scattering. The construction of conventional OCT images is based on lateral scanning of a beam focused within the sample. To increase image acquisition speed and eliminate the need for lateral scanning, wide-field OCT was recently developed. Our experimental and theoretical investigations of the potential and limitations of wide-field OCT revealed the crucial role played by the spatial coherence of the light source. Spatially coherent illumination generates considerable coherent optical cross-talk, which prevents shot-noise-limited detection and diffraction-limited imaging in scattering samples. The dependence on several parameters of the optical system and of the sample properties was investigated in a comprehensive study. Cross-talk increases with the wide-field diameter, numerical aperture, source coherence length, and sample optical density; and strongly depends on sample anisotropy. We showed that spatially incoherent illumination realized with a thermal light source permits cross-talk suppression in wide-field OCT, i.e. rejection of multiply scattered light to a level comparable to that of point scanning OCT. We performed a theoretical study which revealed that the power per spatial mode radiated by thermal light sources is too low to permit a high signal-to-noise ratio while maintaining a fast acquisition speed. Therefore, wide-field OCT realized with either spatially coherent or spatially incoherent illumination suffers from inherent fundamental limitation. This led us to investigate the possibility of exploiting a spatially incoherent light source brighter than a thermal light source. We came to the conclusion that such a "pseudothermal" light source can potentially lead to wide-field OCT systems devoid of cross-talk and an image acquisition speed higher than that of a point scanning OCT system. However, the attractive properties of pseudothermal light sources could be gained at the expense of the simplicity and the economical advantages offered by thermal light sources. Furthermore, fast acquisition speed also relies on a performing "smart pixel detector array". Presently, such detectors do not have sufficient sensitivity and their frequency readout is too low as shown in our feasibility study. We performed a theoretical investigation of the potential of thermal light sources in terms of axial resolution and power per mode. The former revealed that the maximum power per mode is radiated at a wavelength higher than the spectral peak of a blackbody radiator. This led to the important practical conclusion that, at 6000 K, the maximum power is collected in the therapeutic spectral window in OCT (600 - 1300 nm), while at 3000 K this peak is shifted out of the therapeutic window leading to significant power losses. More generally, our work provides a design tool for choosing the optimal thermal light source for a given therapeutic window in terms of signal-to-noise ratio. Currently available sources at 6000 K consist of high pressure gas arc lamps providing a spectrum endowed with spectral lines deleterious for OCT. By suppressing a portion of the spectrum devoid of spectral lines of a mercury arc lamp, we obtained amongst the highest axial resolution so far reported in OCT. Furthermore, the importance of the speckle statistics in OCT incited us to clarify the origin of a difference between two theoretical results reported in the literature. Indeed, two calculations of the amplitude distribution of speckles in OCT, each of them based on a different mathematical formulation, yield different results. We showed that a modification of an initial assumption in one of the formulation leads to equivalent results. In conclusion, this thesis provides a deeper understanding of the potential and limitations of widefield OCT, leading to important design rules. Moreover, it presents a new comprehensive model of OCT putting in question other widely used models.

EPFL2004

Mapping Biological Current Densities With Ultrafast Acoustoelectric Imaging: Application to the Beating Rat Heart

Pierre-Marc Jean Marie Dansette

Ultrafast acoustoelectric imaging (UAI) is a novel method for the mapping of biological current densities, which may improve the diagnosis and monitoring of cardiac activation diseases such as arrhythmias. This paper evaluates the feasibility of performing UAI in beating rat hearts. A previously described system based on a 256-channel ultrasound research platform fitted with a 5-MHz linear array was used for simultaneous UAI, ultrafast B-mode, and electrocardiogram (ECG) recordings. In this paper, rat hearts (n = 4) were retroperfused within a Langendorff isolated heart system. A pair of Ag/Cl electrodes were positioned on the epicardium to simultaneously record ECG and UAI signals for imaging frame rates of up to 1000 Hz and a mechanical index of 1.3. To account for the potential effect of motion on the UAI maps, acquisitions for n = 3 hearts were performed with and without suppression of the mechanical contraction using 2,3-butanedione monoxime. Current densities were detected for all four rats in the region of the atrio-ventricular node, with an average contrast-to-noise ratios of 12. The UAI signals' frequency matched the sinus rhythm, even without mechanical contraction, suggesting that the signals measured correspond to physiological electrical activation. UAI signals appeared at the apex and within the ventricular walls with a delay estimated at 29 ms. Finally, the signals from different electrode positions along the myocardium wall showed the possibility of mapping the electrical activation throughout the heart. These results show the potential of UAI for cardiac activation mapping in vivo and in real time.