Enhanced Blind Speech Signal Dereverberation By Utilizing A Cuckoo-ICA Algorithm [articol]

Abstract

The natural ability of humans to express themselves with their voice has further triggered the development of human-machine interfaces, which allow the control of technical devices using voice commands. Speech dialog systems interact with the user by recognizing and interpreting the meaning of the received commands. The ICA goes to the category of blind source severance (BSS) and the ICA prominently determined by the key assumption of the physical world character. The BSS estimates the original signal using the mixed signal information observed from the input channel. The proposed method avoids the drawback of separated sounds with improper localization, directivity and spatial quality of separate sources. The wavelet filter and multi-step linear prediction coding (mLPC)for extraction of coefficients in the late reverberation. The reverberated signals are eliminated using backward differentiation. Cuckoo is an optimization technique is used to improve the effectiveness of ICA. To reduce the redundant bit and hardware cost the new FPGA was proposed to improve the reliability. Hence the reliability and SNR values are increased. Various methods are compared with proposed Cuckoo search algorithm to demonstrate the efficiency of frequency, time delay and power consumption with reduced area utilization.