Implementation of Spectral Subtraction Using IBM in Simulink

摘要

Spectral subtraction is one of the classical methods for audio de-noising in speech enhancement field, whose purpose is to improve the eminence and perspicuity of speech without any artifact. Spectral subtraction is fundamentally used for background clatter suppression. Diverse calculations have been proposed for speech enhancement, among which the spectral subtraction can be utilized in different ways. This paper portrays the implementation of spectral subtraction in MAT LAB (SIMULINK), using Ideal Binary Masking (IBM), which improves the performance for automatic speech recognition. In this paper Exhibition hall noise with OdB and 5dB are added to clean signal and then enhanced using proposed system. SNR is calculated and compared it with local SNR i.e IBM to estimate the noise and deducted from the noisy signal which ends up within the de-noised signal. Peak Signal to Noise Ratio(PSNR) is used to measure the quality of resulting signal.