LMS-Based Adaptive Filtering Technique for Removing Noise from Voice Signals and Its Comparison with RLS-Based Type (Published)
Every adaptive filter requires an appropriate adaptive algorithm to effectively remove noise components from noise-contaminated desired signals. In line with the requirement finite impulse response filters can be driven by an appropriate adaptive algorithm to remove noise from voice signals. Without the removal, the noise components will degrade the quality of the signal, and the result will be a distortion and substantial loss of message content. In this paper a finite impulse response digital filter driven by least mean square adaptive algorithm for coefficient update is designed to remove overlapping noise component from a voice signal. A real voice statement “Creative Research is Very Essential for the Sustainable Development of Any Nation” is converted to electrical voice signal using microphone and stored in a file in a computer system. With “audioread” command the stored voice signal is loaded into a matlab edit window. The loaded signal is contaminated with a 10.5dB additive white Gaussian noise component generated with matlab. When the contaminated signal is applied to the designed filter the result shows that the noise is effectively removed. The result is evaluated with six properties; listening, signal morphologies, frequency domain analysis, noise attenuation, mean square error, and signal to noise ratio. A comparison of the LMS algorithm and recursive least square algorithm with respect to noise removal from voice signals is performed. The matlab codes for the simulation of this work are provided in this paper.
Keywords: LMS, Voice signals, adaptive algorithm, additive white Gaussian noise
LMS-Based Adaptive Filtering Technique for Removing Noise from Voice Signals and Its Comparison with RLS-Based Type (Published)
Every adaptive filter requires an appropriate adaptive algorithm to effectively remove noise components from noise-contaminated desired signals. In line with the requirement finite impulse response filters can be driven by an appropriate adaptive algorithm to remove noise from voice signals. Without the removal, the noise components will degrade the quality of the signal, and the result will be a distortion and substantial loss of message content. In this paper a finite impulse response digital filter driven by least mean square adaptive algorithm for coefficient update is designed to remove overlapping noise component from a voice signal. A real voice statement “Creative Research is Very Essential for the Sustainable Development of Any Nation” is converted to electrical voice signal using microphone and stored in a file in a computer system. With “audioread” command the stored voice signal is loaded into a matlab edit window. The loaded signal is contaminated with a 10.5dB additive white Gaussian noise component generated with matlab. When the contaminated signal is applied to the designed filter the result shows that the noise is effectively removed. The result is evaluated with six properties; listening, signal morphologies, frequency domain analysis, noise attenuation, mean square error, and signal to noise ratio. A comparison of the LMS algorithm and recursive least square algorithm with respect to noise removal from voice signals is performed. The matlab codes for the simulation of this work are provided in this paper.
Keywords: LMS, Voice signals, adaptive algorithm, additive white Gaussian noise