Noise Reduction Using mel-Scale Spectral Subtraction with Perceptually Defined Subtraction Parameters- A New Scheme

Abstract

The noise signal does not affect uniformly the speech signal over the whole spectrum isn the case of colored noise. In order to deal with speech improvement in such situations a new spectral subtraction algorithm is proposed for reducing colored noise from noise corrupted speech. The spectrum is divided into frequency sub-bands based on a nonlinear multiband bark scale. For each sub-band, the noise corrupted speech power in past and present time frames is compared to statistics of the noise power to improve the determination of the presence or absence of speech. During the subtraction process, a larger proportion of noise is removed from sub-bands that do not contain speech. For sub-bands that contain speech, a function is developed which allows for the removal of less noise during relatively low amplitude speech and more noise during relatively high amplitude speech .Further the performance of the spectral subtraction is improved by formulating process without neglecting the cross correlation between the speech signal and background noise. Residual noise can be masked by exploiting the masking properties of the human auditory system. In the proposed method subtraction parameters are adaptively adjusted using noise masking threshold. A psychoacoustically motivated weighting filter was included to eliminate residual musical noise. Experimental results show that the algorithm removes more colored noise without removing the relatively low amplitude speech at the beginning and ending of words.

Analysis and implementation of low‐power perceptual multiband noise reduction for the hearing aids application

“…With (17) and (18), the VA(n) can be calculated without divisions and logarithms. The second simplification, denoted as the preprocessed input data reuse scheme, exploits the common absolute value of input data at the noise power estimation (9), signal power calculation (12), spectral subtraction (5) and attenuation (6) use the absolute value of input data for processing.…”

“…For such system, many NR methods have been proposed [1,2]. These methods include those based on low-level expansion [3], envelope tracking [4], adaptive low-pass [5] and high-pass filtering [6], envelope modulation filtering [7], spectral sharpening [8], Winner filtering [9,10] and spectral subtraction [11][12][13][14][15][16][17][18]. Among all these methods, spectral subtraction can provide a good tradeoff between quality and computational complexity, which has been a good choice for hearing aid applications.…”

“…Thus, noise attenuation can be adaptive to the noise of each band, and thus results in better speech quality [1]. Based on the mband method, perceptual mband [17, 18] was also proposed by further exploiting perceptual frequency decomposition for efficient noise suppression. However, conventional mband methods cannot be directly applied to hearing aids because of their strict low‐power and low‐latency requirement.…”

“…Targeting for above hearing aids requirement, this paper presents an NR design by exploiting the idea of perceptual mband [17, 18] and entropy‐based VAD [22, 23] as the basis method for further development. This paper is based on our previous publication [24], but provides an in‐depth analysis of algorithm, complexity, speech quality and power consumption as well as detailed architecture and chip implementation.…”