Optimal Algorithms for Blind Source Separation - Application to Acoustic Echo Cancellation.
Masters thesis, National University of Ireland Maynooth.
We are all familiar with the sound which can be viewed as a wave motion in air or other
elastic media. In this case, sound is a stimulus. Sound can also be viewed as an excitation
of the hearing mechanism that results in the perception of sound. The interaction between
the physical properties of sound, and our perception of them, poses delicate and complex
issues. It is this complexity in audio and acoustics that creates such interesting problems.
Acoustic echo is inevitable whenever a speaker is placed near to a microphone in a
general full-duplex communication application. The most common communication
scenario is the hands-free mobile communication kits for a car. For example, the voice
from the loudspeaker is unavoidably picked up by the microphone and transmitted back
to the remote speaker. This makes the remote speaker hear his/her own voice distorted
and delayed by the communication channel or called end to end delay, which is known as
echo. Obviously, the longer the channel delay, the more annoying the echo resulting a
decrease in the perceived quality of the communication service such as VoIP conference
In the thesis, we propose to use different approaches to perform acoustic echo
cancellation. In addition, we exploit the idea of blind source separation (BSS) which can
estimate source signals using only information about their mixtures observed in each
input signal. In addition, we provide a wide theoretical analysis of models and
algorithmic aspects of the widely used adaptive algorithm Least Mean Square (LMS).
We compare these with Non-negative Matrix Factorization (NMF), and their various
extensions and modifications, especially for the purpose of performing AEC by
employing techniques developed for monaural sound source separation.
Repository Staff Only(login required)
||Item control page