Musical Noise Reduction Capability of Various Speech Enhancement Algorithms
Prateek Saxena, Vijay Kumar Gupta, Mahesh Chandra · Advances in intelligent systems and computing · 2016
This paper presents a comparative analysis of spectral subtraction and Weiner denoising techniques for musical noise reduction. The iterative spectral subtraction method provides least musical noise generation applied in different noisy environments. The method of musical noise production is traced by observing the change in the kurtosis ratio of noise spectrum using different denoising techniques for different noisy signal. A MATLAB simulation is performed for four different noisy environments car noise, babble noise, operation room noise and machine gun noise at −10, −5, 0, 5 and 10 dB input SNR levels. It is observed that wiener based methods provide more improvement in SNR as compared to spectral subtraction based methods. But at the same time musical noise generation is more in wiener based methods. The wiener based method HRNR gives a maximum 35.77 dB improvement in SNR for car noise at −10 dB input SNR level. Iterative spectral subtraction gives the minimum value of kurtosis ratio for all noises at all input SNR level.