|
The following is a mostly-complete listing of published research since
the graduate music engineering program began in 1986. They are listed
in reverse chronological order - highlighting the graduate student author
and the topic of their thesis. Both the thesis abstract and the
entire published thesis are available (for those we have electronic copies
for). If you are interested in a simpler chronological listing of our
research work, you should click on the "Graduate Thesis"
link to the left. 2003 - Robert Hartman Spatially Relocated Frequencies and Their Effect on the Localization of a Stereo ImageThe ability of humans to detect the location of a sound is generally referred to as localization. Sound waves interact with the head, body, and pinnae creating temporal and spectral differences between the left and right ear canal signals. The brain uses these differences to interpret a probable number of sound events and their respective locations. There are three major cues: interaural time differences (arrival, phase, envelope), interaural level differences, and the monaural pinnae influences. The physical presence of these cues depends mostly on the spectral content of the sound and its spatial origin relative to the listener. Perceptually, the localization cues exhibit a relative dominance that varies significantly with frequency. This research explores the relative importance of low and high frequency localization cues during free field listening. More specifically, it compares the horizontal shift of a stereo image caused by spatially relocating low versus high frequency bands of the audible spectrum. It is shown that contrary to the popular belief that low frequencies are “hard to localize,” the horizontal position of a stereo image is most significantly affected by moving low-to-mid frequencies as opposed to high frequencies. This can most likely be attributed to the overall perceptual dominance of low frequency interaural phase differences. top 2002 - Eduardo Trama Sound Source Sepaeration using 3-D Correlogram, Fuzzy Logic, and Neural NetworksWith the advance of neuroscience, new mathematical tools were developed to help engineers build computational intelligent systems that simulate the way the human brain processes external information. The problem of separating one sound signal from a sound mixture is based on how we extract the cues necessary to distinguish that particular signal. When two related nonlinear sound signals are mixed together (composition) it is impossible to separate (decomposition) them with traditional linear mathematical methods. The human brain can easily distinguish two different persons talking at the same time. It uses its previous knowledge of at least one of the sound signals. In this project an Artificial Fuzzy-Neural System will be used as the main tool to detect and classify a sound library database. A database will be used for sound search and analysis in order to separate two or more sound signals from a mixture. To preprocess the sound mixture a 3D Correlogram will be used as the auditory scene analysis tool. The final approach of this research is to design a system that learns every new sound signal that will be used with the next sound separation. top 2002 - Rodrigo Ordonez The Effects of Cross-Modal Interaction in the Perception of Audiovisual Quality and its Application in Data Reduction AlgorithmsModern lossy audio data reduction algorithms such as MPEG-1, MP3Pro, WMA and AAC are based on the perceptual characteristics of the listener in response to auditory signals. Data reduction is achieved by discarding, or selectively quantizing, information that will most likely not be perceived by the listener. These same methods can be directly adopted for use in audiovisual applications in which the viewer receives simultaneous information through multiple senses. This paper presents an analysis of the effects of cross-modal interaction between temporally coherent visual and auditory signals in the perception of audiovisual quality for compressed audio. It suggests the framework for a feature extraction and prediction model that could be applied towards more efficient audiovisual data reduction algorithms. Experimental tests are reported to verify the validity of the model. top 2001 - Margarita Escobar Beat Detection in Music Using Average Mutual InformationIn this research an alternative automatic beat detection method using average mutual information (AMI) is proposed and tested on music audio files. This method carries on envelope detection, decimation, and applies AMI (a non-linear correlation). AMI is a measure of how much information one can expect to derive from future measures of a system based on current observations. In this particular study, AMI is applied to music files to obtain the beat rate due to the dependence in the signal is assumed to be the beat. Experimental results show that the method is effective and could be a good option in applications where real-time are needed even though the AMI calculation could be demanding of computational resources. Some audio files were used for testing such as: train of pulses, popular music sampled from compact discs, and MIDI files. All files had constant beat rates. The system performed better with frames of 5 sec, fs= 8 kHz, and with music files having beats clearly marked by percussion. The system presents some capacity to detect strong and weak beats (quarter note level), this opens opportunities in the development of algorithms capable of detecting levels of rhythmic structure in music. In addition, this study is relevant for those interested in beat detection and beat tracking. top 2001 - Kim Hang Lau A System for Hybridizing Vocal Performance (entire thesis: HTML format or PDF Format - Coming soon) Signal processing for the signing voice is a relatively new and interesting field. Unlike singing synthesis which aims to recreate the human voice, vocal modification aims to modify live or recorded vocal parts by manipulating its timbre, pitch contour, timing patterns and/or amplitude envelope. A prototype system for vocal modification was developed. This system modifies a source vocal sample to match the time evolution, pitch contour and amplitude envelope of a similarly sung target vocal sample. Musically, this system simulates a non-parametric transfer of singing techniques from the target vocalist to the source vocalist. The system is comprised of a time-varying time/pitch/amplitude modification engine, a pitch-detector and a subsystem for the generation of modification parameters. Further improvements on the individual components are recommended to handle greater dynamic changes of the vocal signal, thereby extending the current good results for slow and continuous singing to a wider range of singing styles. top 2001 - Sean Browne Hybrid Reverberation Algorithm Using Truncated Impulse Response Convolution and Recursive Filtering (entire thesis: HTML or PDF format) Artificial reverberation can be developed by convolving an input signal with the specific impulse response of an acoustic space. This computationally exhaustive method achieves superb reverberation, but is often abandoned for simpler filtering methods. Since it is only the length of the impulse response that determines the computational cost and resolution, a new hybrid algorithm is developed that borrows from the original impulse response easing the computational burden. A truncated impulse response can still convey the rich energy of the space while a digital filter bank can model the late echoes and reverberation tail. This more efficient algorithm will use a reduced length block convolution and recursive filter network to achieve similar high quality reverberation. top 2000 - Jasmin Frenette Reducing Artificial Reverberation Requirements using Time-variant Feedback Delay Networks (entire thesis)(HTML format)Most of the recently published artificial reverberation algorithms rely on a time-invariant feedback delay network (FDN) to generate their late reverberation. To achieve a high-quality reverberation algorithm, the FDN order must be quite high, requiring a good amount of memory storage and processing power. However, in applications where memory and computational resources are limited such as hardware synthesizers or gaming platforms, it is desirable to achieve a good sounding reverberation. This thesis proposes the use of time-variant delay lengths to maintain the quality of the reverberation tail of an FDN, which reduces the algorithm’s processing time and memory requirements. Several modulators are evaluated in combination with several interpolation types for fractional delay interpolation. Finally, the computation efficiency and memory usage of the time-variant reverberation algorithms are compared with the equivalent quality, higher order, time-invariant algorithms. top 2000 - Abhijeet Tambe An Adaptive Time-Frequency Representation with Re-synthesis using Spectral Interpolation (entire thesis)(Adobe *.PDF format)The conventional method for storing and transmitting digital audio is with PCM samples in the time domain. An adaptive time-frequency analysis is performed on the audio signal and it is stored in the form of a 3-D matrix containing time, frequency and magnitude information. The validity of this method is tested by using a spectral interpolation algorithm to perform re-synthesis in order to compare the synthesized signal to the original. Among the many applications for such a representation, the possibility of audio compression is explored. The results of applying this procedure on a few audio signals are plotted along with the results of listening tests. top 2000 Daisuke Koya Aural Phase Distortion Detection (entire thesis)(HTML format)Previous research has proven that phase distortion in audio signals is audible. The real question then is not the existence, but the significance of the audibility. A psychoacoustic experiment is proposed to ascertain permissible levels of phase distortion in audio signals. This thesis research is not meant to be exhaustive by any means, but to ascertain some permissible levels based on careful experimental design and analysis. The Kwalwasser-Dykema Music Tests format will be used in the implementation of the thesis research with other considerations. These permissible levels may be beneficial in the design and application of audio equipment, especially in the area of transducer and loudspeaker system engineering. top 2000 Alexander Iliev Unavailable, Author - please contact webmaster to provide thesis information top 1999 - Justo A. Gutierrez Duffing's Equation as an Excitation for Guitar Models (entire thesis)(HTML format)Many musical instruments can be described by a linear resonator that is excited by a nonlinear oscillator. It is this nonlinear oscillator that provides much of the characteristics for a given instrument. The primary objective of this study was to provide the basis for a new excitation mechanism for plucked string instrument models which utilizes the classical nonlinear system described in Duffing’s Equation. Guitar samples were inverse filtered through a general string model to obtain a residual. This provided a starting point from which parameters of a numerical solution for Duffing’s Equation were then manipulated to produce an approximation of this residual. A general method of setting these parameters was proposed for producing approximations of a desired waveform. These signals were then fed back into the general string model. The synthesized guitar could resemble the original sample, and this hence proved that Duffing’s Equation could indeed be used as a starting point for creating a user-controllable excitation for the physical model of a plucked string instrument. Furthermore, Duffing’s Equation can also be manipulated to form a variety of waveforms which may be useful to musicians in all forms of music synthesis. One of the main benefits of this system is that it is fast and it is something that can be implemented without the need for memory storage of instrument samples. top 1999 - Alex Souppa Minimum Weighted Norm Extrapolation of Digital Audio Using Frequency Domain Blocking (entire thesis)(HTML format)When faced with the absence of digital audio data due to such problems as packet loss during real-time Internet streaming or defective media, it is often desirable to replace the missing data with relevant information. Interpolation can be used if the missing data segment is short enough and if known uncorrupted data surrounds the missing data. If, however, the missing data is too large for valid interpolation and/or succeeding data is unknown, extrapolation could be used. This thesis proposes a new minimum weighted norm extrapolation algorithm for digital audio data using frequency domain blocking. Due to the high sampling rates commonly used in digital audio, small time segments are comprised of a large number of samples. The frequency-domain blocking method is included in the algorithm to deal with this sample-intensive nature of digital audio. This method reduces the computational requirements and enables the extrapolation of larger data segments by dividing the known time-domain data segment into several smaller time-domain data segments. Each smaller segment is processed through the extrapolation method and then re-combined to form the main extrapolation. The algorithm has no bandwidth limitations, and it is intended to provide the listener with a continuous audio stream even in the presence of missing audio data. Listening tests are performed and analyzed evaluating the perceptual quality of the algorithm. These tests demonstrate good perceptual quality of the algorithm filling gaps of various lengths from 10ms to 640ms in audio examples of different musical styles. top 1999 - Ricardo Garcia Digital Watermarking of Audio Signals Using a Psychoacoustic Auditory Model and Spread Spectrum Theory (entire thesis)(HTML format)A new algorithm for embedding a digital watermark into an audio signal is proposed. It uses spread spectrum theory to generate a watermark resistant to different removal attempts, and a psychoacoustic auditory model to shape and embed the watermark into the audio signal while retaining its perceptual quality. A software system is implemented and tested for perceptual transparency and data-recovery performance. top 1999 Jason Flaks Unavailable, Author - please contact webmaster to provide thesis information top 1999 Miguel Chavez Unavailable, Author - please contact webmaster to provide thesis information top 1999 David Alonso-Dominguez Unavailable, Author - please contact webmaster to provide thesis information top 1999 Stephen Handley Unavailable, Author - please contact webmaster to provide thesis information top 1998 - Jason VandeKieft Computational Improvements to Linear Convolution With Multi-rate Filtering Methods (entire thesis)(HTML format)Possible reductions in the computational complexity of linear convolution are explored. Specifically, multirate filter banks are implemented in such a way as to reduce the complexity of the direct form linear convolution computation by nearly half. The proposed multirate approach is not ideal. Some aliasing in the convolution result occurs due to the reduced computational load. Listening tests with musical input material and room impulse responses show that this distortion is audible for some types of input, but imply that the differences are not musically significant. One pre-existing multirate algorithm (Vetterli, 1988) may outperform well-known fast convolution techniques in the frequency domain. top 1998 - James R. West Five-Channel Panning Laws : An Analytical and Experimental Comparison (entire thesis)(HTML format)Panning is the technique of adjusting the perceived horizontal azimuth of a sound produced by two or more loudspeakers. Five panning methods employing interaural intensity difference (IID) cues are described for application in creating 5.1 channel surround sound recordings for domestic reproduction. The first four methods are optimized for constant gain, constant power, velocity and energy vector equality, and avoidance of azimuthal aliasing, respectively. A fifth algorithm is developed as a compromise between the constant power and vector optimizations. All five algorithms are compared analytically using the four optimization techniques. Listening tests are conducted on all but the constant gain algorithm. They are necessary because the perceptual significance of the vector and azimuthal aliasing optimizations is unknown. Tests using stationary and moving pans are conducted with 11 subjects in a non-anechoic room. Measurements are taken of the accuracy of horizontal azimuth, consistency of speed, consistency of distance, and consistency of image width. Results indicate that vector optimized panning is superior for center listening and constant power panning is superior for off-center listening. It is unclear if vector optimized panning performs well because of its optimization method or other factors. The constant power and vector optimized algorithms are implemented in C++ as a DirectX audio plug-in for Windows95. top 1998 Mathew Abraham Unavailable, Author - please contact webmaster to provide thesis information top 1998 Ali Habashi Unavailable, Author - please contact webmaster to provide thesis information top 1997 - Matthew VanDyke Kotvis An Adaptive Time-Frequency Distribution with Applications for Audio Signal Separation (entire thesis)(Adobe *.PDF format) An adaptive time-frequency distribution is developed to represent an audio signal in a more useful manner. The intended use of the adaptive time-frequency distribution is to aid digital signal processing systems in identifying specific components in a signal. This distribution has applications to many facets of digital audio signal processing, including lossy audio compression and signal separation. As an example, the specific application of removing sources from two channel audio recordings is investigated. top 1997 - Kirk Lampert Methods to Reduce the Bandwidth Requirements of MPEG-1 Layer II Audio Data for Transmission Speeds of Less Than 28.8 kbps (entire thesis)(HTML format)MPEG-1 Layer II offers standardized audio compression but does not provide data rates low enough to allow audio streaming over the Internet through low bandwidth modems. Methods are introduced in this project to reduce the bandwidth requirements of MPEG-1 Layer II audio data. New tools are developed to implement these methods and comparisons are made between the new very low bit rate files, RealAudio 3.0, and I-Wave. Pre and post processing methods are also discussed. top 1995 - John Antony Adaptive Techniques to Reduce Quantization Error in an MPEG-1 Video EncoderSeveral adaptive algorithms are developed to reduce quantization error in an MPEG-1 video encoder. Advantages and disadvantages of traditional and experimental image error measures are addressed as their use in adapting an encoder's quantization scale factor is explored. Results of the different adaption techniques are evaluated both quantitatively and qualitatively. top 1995 - Mauricio Greene Effects of Minimum-Phase HRTFs on a Cross Canceling FilterSoftware of three dimensional auditory display is developed. Monaural sounds are converted into binaural format for headphone listening and then to a transaural format for loudspeaker listening. The principal sound localization cues of measured head related transfer functions (HRTF) are analyzed. Transaural format is achieved through the implementation of a cross- canceling filter. A transaural acoustic matrix is laid out and a cross-canceling filter is derived from it. It has been found that head related transfer functions can not be used directly as cross-canceling filter coefficients since they will generate an unstable system, HRTFs need to be transformed into their minimum- phase version to maintain stability in the system; HRTFs are transformed to minimum-phase HRTFs using a cepstrum algorithm. Furthermore, interaural time information is lost in the minimum- phase conversion and needs to be reinserted to have suitable HRTF for adequate 3-D audio rendering. Finally minimum-phase HRTF effects are analyzed. Result and conclusions are drawn with the proper analysis. top 1995 - Aurika Checinska Hays Pitch Recognition and Intonation Error Quantification of Discrete Violin TonesMusical pitch recognition is explored from an engineering standpoint and quantitative intonation error assessment is implemented. Violin tones are chosen due to their complex spectral characteristics which, as in speech, can vary greatly form instrument to instrument. The system, implemented under Microsoft Windows, samples 16-bit monophonic audio from a microphone. The data is transformed into the frequency domain via an FFT. The spectrum is then processed by one of several recognition algorithms, resulting in a closest pitch match. A resolution enhancement procedure is performed to identify the exact frequency played and to determine an intonation error percentage. Various recognition algorithms were developed and tested, the most successful of which integrated the effects of instruments formants and periodicity pitch. top 1995 - Luis Martinez An Efficient ISO/MPEG-1 Layer II Encoder Using Parallel Processing TechniquesA sequential analysis of the different sections of an ISO/MPEG-1 layer II encoder is provided. From the analysis, the filter bank and FFT analysis sections of the encoder are identified as the most time consuming segments of the coding process. Two parallel structures and mapping techniques for a more efficient implementation of both sections of the encoder are described. The topologies are based on mesh-connected and cubic networks. Details on the realization and time requirements of the structures are provided, as well as comparisons from the results of the performance analysis. A parallel implementation of the filter bank section of the encoder, using the Occam parallel programming language, is also presented, and the outcome of the programming implementation is validated. The results show that the parallel encoder performs significantly better than the sequential encoder within certain parameters of both implementations. The project encourages the pursuit of more parallel realizations of other sections of the MPEG-1 layer II coding algorithm, such as the bit allocation section, that diminish the overall performance of the proposed parallel implementation. top 1994 - Matthew Fellers Audio Compression for Wavetable SynthesisAn algorithm is developed for adapting channel-based perceptual coding techniques to Wavetable synthesis. The need for pitch transposition of perceptually coded waveforms introduces new requirements on a psychoacoustical model. This coder draws upon Psychoacoustical Model 1, outlined in the ISO/MPEG-1 audio Standard 11172-3 as a starting point, and makes key modifications to assure coding artifacts are minimal upon pitch transposition. Allocation of bits to bins relies on a psychoacoustical model in the frequency range from DC - 7 KHz while vector quantization (VQ) is used for coding in the 7 - 22 KHz range. The coding of digitized musical instrument waveforms in the frequency domain not only complements data reduction, but also offers new possibilities for timbral control, either in real time or non- real time. top 1994 - Frank Filipanits Optimization Techniques to Reduce Temporal Computation Requirements of Auralization Processing Algorithms (entire thesis)(HTML format) An auralization software system for performing static (non- moving) sound source placement with headphone playback is developed using present theory and algorithms. Several caveats for auralization system design are identified and addressed. One method of temporal computation optimization is then presented. It is shown that bandwidth analysis of the raw sound source greatly reduces the computation time necessary for auralization synthesis, by identifying frequency ranges which contain zero information and can be ignored during processing. a residue method is used to evaluate the resulting algorithm. top 1994 - Timothy Onders An Object-Oriented Software Codec for Evaluation of MPEG-1 Audio Coding SystemsPsychoacoustic-based data reduction techniques have become a key component of many new media systems. In order to provide the most flexible coding system, the MPEG-1 audio standard only specifies the decoding process. An object-oriented MPEG-1 software codec has been developed to provide a design framework for MPEG-1 audio coders and other related systems. The object- oriented design has the added benefits over previous implementations of improved transportability, maintainability, and an inherently parallel design for use with the next generation of DSP processors. top 1993 - Glenn Josefiak Use of a Neural Network for Time-Domain Estimation of Formants in SpeechA two-layer neural network is designed for the task of formant tracking in continuous natural speech. It operates on time- domain speech samples, eliminating the need for frequency transformation. The formant values used for training data are derived by modeling the speech waveform as the sum of damped sinusoids. The optimal frequencies, bandwidths, and amplitudes are determined by the minimization of a least-squares error function between the original speech and the approximation. Results show that the time-domain sinusoidal model is practical for the formant tracking problem. The neural network is also shown to be capable of learning a transfer function which can reproduce the parameters generated by this model. While the neural is not yet capable in all cases of accurately tracking formants, in one instance where the net was tested on the training patterns, it correctly classified formant presence more than 75% of the time and correctly estimated formant frequency more than 90% of the time for each of the first four formants. This is encouraging for future research. top 1993 - Thomas Zudock Effects of Minimizing Impulse Response Duration in Three Dimensional Sound ProcessingSimulation of binaural recordings was accomplished through head related transfer function measurements and room modeling techniques. Head related impulse responses were obtained through measurements from a dummy head. Reverberant impulse responses were generated based on the early reflections indicated by ray tracing data from an acoustical modeling software package. Multiple cases concerning the number of reflections used in an impulse response were developed to simulate different distances and determine if externalized sound images may be produced with a fairly short impulse response. The four cases included dry (no reflections), first order reflections, second order reflections, and third order reflections. Subjective evaluation suggested that first order reflections provided enough information to externalize processed sound for nearly all cases while dry stimuli in many cases did not. Additionally, increasing the number of reflections continually increased perceived distance for some subjects (at least up to the maximum number of reflections used in this experiment) while for other subjects it eventually caused a decrease in perceived distance. The greatest perceptual errors concerned a tendency to lateralize the position of the sound images and to internalize simulated positions immediately in front of or behind the listener. top 1993 - William Johnston A Prototype Expert System for Jazz HarmonizationA prototype expert system was developed for the Macintosh computer, using Pascal as an implementation language. The system accepts a given sequence of chords and a melody, and constructs a four or five note harmonization of the melody based on the chords. Analysis of the melody is performed to ensure a good correlation between the chords and the melody. Chords entered are modified accordingly, and the harmonization is performed. The system utilizes a typical expert system architecture including a knowledge base, and inference machine, and a user interface, and is useful for both the novice and skilled user. Arguments are made for procedural versus heuristic methods, and suggestions for future additions to the system are presented. top 1992 - Jayant Datta Software to Analyze the Performance of Windows with Respect to the Auditory Masking Curve Low bit rate coding of audio signals is an area of increasing importance. Very briefly, the time domain input signal is transported to the frequency domain by means of a fast transform algorithm. A model of the ear and other psychoacoustical parameters are used to determine the perceptibility of the spectral components of the input signal. The masked components are not coded; resulting in data reduction. Before entry into the frequency domain, it is necessary to window the input signal. It turns out that the spectral analysis results are influenced by the type of window used. A program has been written to model the ear and compare the performance of a window with the theoretical response of the ear. Windows are judged on their ability to keep the windowed response under the auditory masking curve. top 1992 - Robert Dunn Digital Signal Processing Algorithms for Noise Reduction in Digital AudioThis project involves the application of digital signal processing theory to the design of a functioning software tool for digital audio. Standard linear filters, a nonlinear median filter, and a frequency-domain adaptive filter were designed and implemented as part of the project. As well as requiring the understanding and application of the theory required, the project required the use of several sophisticated tools for software development. The resulting algorithms can be used on compact disc quality digital audio to do pre-emphasis and de-emphasis filtering, DC offset removal, median filtering, and adaptive noise canceling. top 1992 - Ted Tanner Subband: A Software Simulator for Analysis-Synthesis FilterbanksThis paper will discuss the theoretical as well as practical aspects of implementing a Subband filter bank algorithm using band pass window FIR filters. This is to simplify the design process and reduce the computational complexities involved with most of the current filter bank schemes. The paper will discuss the theoretical basis for simplification of the subband filter bank design in the format of psychoacoustical foundations for current findings in audiology, as well as the thoroughly researched areas of critical bandwidth. In addition to the critical bandwidth of the ear, non-linearities will also be explained for applications in more efficient coding schemes. The filter bank is a multi-channel system that implements these critical bandwidths into actual band-pass window finite impulse response (FIR) filters and is able to process 16 bit linear PCM audio information. The filter bank will operate on the basis of increased threshold shifts (ITS) that was found to be prevalent in our society. The environment of the software will be discussed in detail as to provide for future upgrades and improvements. top 1992 - Brent Karley Modular Program for Automating the Alignment System for the APR-24 Multitrack Tape RecorderFrequent alignment of audio tape recorders is necessary to maintain optimal operating conditions. This is typically a complicated and tedious process which consumes much of a broadcast or recording engineer's time. several manufacturers of audio tape recorders have expedited this process by utilizing digital technology. This has resulted in the design of tape recorders which provide digitally adjustable alignment parameters utilizing digital to analog converters (DACs), rather than the conventional analog adjustments using potentiometers. These tape recorders are equipped with a serial interface (typically RS-232 or RS-422) allowing for external control of the alignment parameters. Personal computer-based audio test equipment is available to make appropriate measurements in the alignment process. A software Program has been developed to operate on a personal computer to control both an audio tape machine and an audio test system to produce automatic alignments. This program also provides for storage of alignment parameters to disk or hardcopy. By utilizing the automation the computer can take over the alignment process allowing the engineer to perform other tasks. The alignment is thus performed in a fast and accurate manner and accommodates quick repeatability with its data saving feature. The simplicity of the automated alignment system allows less experienced engineers to perform the alignment and provide a printed summary of all alignment parameters. top 1991 Daniel Mikat Unavailable, Author - please contact webmaster to provide thesis information top 1991 Sean Stevens Unavailable, Author - please contact webmaster to provide thesis information top 1990 - Michael Ballman Auditory Localization Research TooResearch in the area of auditory localization has just come into prominence this past decade. Musicians, recording engineers and producers, audio equipment designers, acousticians, physicians, psychologists, and even military weaponry designers have acquired the need and desire to study all the facets and limits of human ability to localize sound. In particular, each application stresses the importance of learning how certain parameters of a sound affect localization sensitivity. After becoming familiar with much of the research in this area, it became apparent that almost all of the studies in this area examine the effects of just one variable (or two or three at the most) on human sensitivity to sound localization. It is difficult to draw solid conclusions with this type of procedure because there are several parameters which are believed to interactively affect sensitivity to auditory localization. In addition, it seems that much of this research is still being conducted without the assistance of computer-driven test systems. With this in mind, this project focused on the development of a powerful computer-aided system which would facilitate the study of how numerous cues interact with each other and influence sound localization sensitivity. The flexible and user-friendly Auditory Localization Research Tool would allow for one, all, or any number of parameters to be varied so that a particular localization test could be designed for any situation. top 1989 - Marc Bavay Adaptive FIR Filtering in Time and Transform DomainsThe availability of an increasing number of digital audio recordings has encouraged the development of all-digital audio systems where music would stay from the recording to the mastering processes under it s digital format. Audio signal processing can largely benefit from adaptive signal processing techniques such as adaptive interference cancellation, unknown system modeling and equalization. The goal of this research project is to present a time- domain adaptive finite impulse response (FIR) filter implemented as a tapped (or weighted) delay line using the popular Least Mean Square (LMS) algorithm to adapt its own coefficients (or weights), for real and complex input signals, and to derive a frequency-domain implementation of this filer, using different orthogonal transforms. The behavior of the two implementations are compared, using simulations where the filter tries to recover a sine wave embedded in white noise with a signal-to- noise ratio of 10 dB. It is found that for complex input signals, the optimum Wiener weight solution for the coefficients of the filter is different that the one stated in the literature. It is demonstrated by theory and simulations that the time and frequency-domain implementations have the same behavior provided there is a condition on the convergence factor of the LMS algorithm. Finally, applications of adaptive techniques to the processing of audio signals are discussed. It opens new possibilities, where devices such as loudspeakers are able to correct their own aberrations, and automatically equalize non- ideal drive unit response. top 1989 Chin-Woo Park Unavailable, Author - please contact webmaster to provide thesis information top 1989 Kevin Ryan Unavailable, Author - please contact webmaster to provide thesis information top 1989 Randall Lopez Unavailable, Author - please contact webmaster to provide thesis information top
|