Efficient De-noising Performance of a Combined Algorithm of Translation Invariant (TI) Wavelets and Independent Component Analysis over TI Wavelets for Speech-Auditory Brainstem Responses

Abstract

In this paper we have presented a research for de-noising the EEG collected Brainstem Speech Evoked Potentials data collected in an audiology lab in University of Ottawa, from 10 different human subjects. Here the de-noising techniques we have considered are Yule-Walker Multiband Filter, Cascaded Yule-Walker-Comb Filter, Conventional Wavelet Transform estimation filters: Daubechies, Symlet, Coiflet Wavelet families, Translation Invariant (TI) Wavelet Transform estimation filter, FAST Independent Component Analysis (FASTICA) De-noising Technique, Combined algorithm of “Translation Invariant (TI) Wavelets and Independent Component Analysis” De-noising technique. The performance measures we have considered are Mean Square Error (MSE) and Signal-to-Noise-Ratio (SNR) values. Out of these techniques we found that cascading of Yule-Walker filter and Comb-Peak filter gave better De-noising performance than Yule-Walker Multiband Filter. Then conventional Wavelets performed far better than the cascaded filter, in those Daubechies family of wavelets worked better than all. Then FASTICA Algorithm worked near to the performance of Conventional Wavelets but far better than cascaded filter. Then we have utilized Translation Invariant (TI) wavelet algorithm which provided the excellent performance than above all. Then we have utilized combined Algorithm of “Translation Invariant (TI) Wavelets and Independent Component Analysis – CSTIICA” algorithm which found to be, it may perform better than TI wavelets algorithm. Ultimately TI and CSTIICA algorithms are found to be may be the best auditory artifact removal techniques and can be highly useful in auditory EEG data analysis to the best.