A COMPARATIVE ANALYSIS OF INDEPENDENT VECTOR AND COMPONENT ANALYSIS IN TERMS OF PERFORMANCE EVALUATION

Abstract

In accordance with the energy of the no-stationary mixed signal the weights of the hybrid model
between Gaussian and Gaussian will be allocated. On the other hand, in many practical areas of biomedical
and engineering, ICA is a newest idea in the statistics and extensively utilized approach for the purposes of
BSS. EEG and ECG are multi-channel recordings which represent bodily activities. These multi-channel
recording are particularly difficult to understand because of the complicated propagation feature of human
tissue. The various methods of the ICA, however, extract signals that may be easily associated with specific
bodily function. It was based on a non-Gaussanity technique to discover independent sources, are based on
an advanced version of the fast ICA method of Aapo Hyvärinen and Erkki Oja. MATLAB methods have
been created based on linear IVA and ICA mathematical models in this thesis. These methods are used to
identify the de-mixing matrix for the signal mixture, thereby isolating the words of each source. Laplacian
distribution capabilities mean that speech signals in themselves are leptokurtic such that both IVA and ICA
could be recognized easily. Subjective and objective quality tests were used to evaluate the increased signals
from both IVA and ICA. The average signal-to-noise ratio (SNR) result and mean opinion indicate that the
ICA technique is better suited for this task.