Date of Award


Document Type


Degree Name


Organizational Unit

Daniel Felix Ritchie School of Engineering and Computer Science, Electrical and Computer Engineering

First Advisor

Kimon P. Valavanis

Second Advisor

Wenzhong (David) Gao

Third Advisor

Mohammad Abdul-Matin

Fourth Advisor

Frederic Latremoliere


Block-PCA, MMSE filtering, Noise reduction, Principal component analysis, Speech and audio enhancement


Smart and intelligent devices are being integrated more and more into day-to-day life to perform a multitude of tasks. These tasks include, but are not limited to, job automation, smart utility management, etc., with the aim to improve quality of life and to make normal day-to-day chores as effortless as possible. These smart devices may or may not be connected to the internet to accomplish tasks. Additionally, human-machine interaction with such devices may be touch-screen based or based on voice commands. To understand and act upon received voice commands, these devices require to enhance and distinguish the (clean) speech signal from the recorded noisy signal (that is contaminated by interference and background noise). The enhanced speech signal is then analyzed locally or in cloud to extract the command. This speech enhancement task may effectively be achieved if the number of recording microphones is large. But incorporating many microphones is only possible in large and expensive devices. With multiple microphones present, the computational complexity of speech enhancement algorithms is high, along with its power consumption requirements. However, if the device under consideration is small with limited power and computational capabilities, having multiple microphones is not possible. For example, hearing aids and cochlear implant devices. Thus, most of these devices have been developed with a single microphone. As a result of this handicap, developing a speech enhancement algorithm for assisted learning devices with a single microphone, while keeping computational complexity and power consumption of the said algorithm low, is a challenging problem. There has been considerable research to solve this problem with good speech enhancement performance. However, most real-time speech enhancement algorithms lose their effectiveness if the level of noise present in the recorded speech is high. This dissertation deals with this problem, i.e., the objective is to develop a method that enhances performance by reducing the input signal noise level. To this end, it is proposed to include a pre-processing step before applying speech enhancement algorithms. This pre-processing performs noise suppression in the transformed domain by generating an approximation of the noisy signals’ short-time Fourier transform. The approximated signal with improved input signal to noise ratio is then used by other speech enhancement algorithms to recover the underlying clean signal. This approximation is performed by using the proposed Block-Principal Component Analysis (Block-PCA) algorithm. To illustrate efficacy of the methodology, a detailed performance analysis under multiple noise types and noise levels is followed, which demonstrates that the inclusion of the pre-processing step improves considerably the performance of speech enhancement algorithms when compared to other approaches with no pre-processing steps.

Publication Statement

Copyright is held by the author. User is responsible for all copyright compliance.

Rights Holder

Abdullah Zaini Alsheibi


Received from ProQuest

File Format




File Size

116 pgs


Electrical engineering, Computer engineering