Noise Reduction in Speech Signals Using Recursive Least Square Adaptive Algorithm (Published)
In many types of communications that involve speech systems it has been observed that the speech signals are easily interfered with by noise to corrupt it and this tampers with the output accuracy and performance of the systems. It is important that the noise is filtered out from the corrupt speech signal to improve the performance and accuracy of the speech systems or quality of the speech for listening. Filtering out the noise in practice without compromising the integrity of the speech signal is not very simple. Several speech filtering algorithms have been used by various researchers to filter out noise present in speech. In this paper a finite impulse response adaptive filter of order 32 and forgetting factor of λ=1.0, based on recursive least square adaptive algorithm for coefficient update is designed to filter out additive white Gaussian noise from a speech signal. The speech signal is produced by converting a real voice statement “Recursive Least Square Adaptive Algorithm is Very Efficient in the Processing of Speech Signals” to speech signal with a microphone and stored in a file in a computer system. An “audioread” command is used to load the stored speech signal into a matlab edit window and a 10.5db additive white Gaussian noise is generated with matlab and used to corrupt the loaded speech signal. Applying the corrupt speech signal to the designed filter indicates that the noise is drastically reduced. We used four properties to evaluate the performance of this algorithm and they are the output sound, signal morphology, frequency distribution and the filter attenuation strength. The matlab program code for the simulation is provided in this paper.
Keywords: Power spectral density., RLS, adaptive algorithm, additive white Gaussian noise, speech signal