Window technique is widely used in speech processing to enhance the quality of speech by using it in the design of FIR filters for removing or reducing noise components associated with speech signal propagation or transmission. Such noise components include Additive White Gaussian Noise (AWGN), Random Noise, power line noise, high and low frequency noise components. The information content of any contaminated speech signal is unreliable unless these noise components are removed. This paper presents a FIR filter designed with double- cosine term generalised adjustable window with to denoise speech signal of high frequency noise components. The speech signal is recorded as windows media audio (.wma) format and as such the optimal sampling frequency is 44100Hz while the filter order is 34. A real voice statement, “Education is the Key to the Development of any Nation” is transduced to electrical form with the in- built microphone of a laptop system, recorded in windows media audio (.wma) format and stored in a file in the system. With “audioread” instruction the stored voice is loaded into a matlab work space. A noise component of 4500Hz and above is generated with matlab and mixed with the speech to obtain a contaminated speech signal. The designed filter is able to effectively denoise the contaminated speech signal when it is applied to it.

Keywords : Adjustable Window, Double-Cosine Term, Speech Signal, High Frequency Noise, Power Spectral Density.