A Noise Estimator with Rapid Adaptation in Variable-Level Noisy Environments

A Noise Estimator with Rapid Adaptation in Variable-Level Noisy Environments Bing-Fei Wu, Kun-Ching Wang*, Lung-Yi Kuo Department of Electrical and Control Engineering National Chiao-Tung University Hsinchu, Taiwan, R.O.C. Corresponding author mail: Kunching@cssp.cn.nctu.edu.tw Abstract. In this paper, a noise estimator with rapid adaptation in a variable-level noisy environment is presented. To make noise estimation adapt quickly to highly non-stationary noise environments, a robust voice activity detector (VAD) is utilized in this paper and it depends on the variation of the spectral energy not on the amount of that. The noise power spectrum in subbands are estimated by averaging past spectral power values using a time and frequency dependent smoothing parameter, which is chosen as a sigmoid function changing with speech-present probability in subbands. The speech-present probability is determined by computing the ratio of the noisy speech power spectrum to its local minimum. Noise measurement, speech enhancement, spectral analysis, signal process. 1 Introduction An accurate noise estimator used for speech enhancement in adverse environments is one of most essential parts. Inaccurate noise estimator will result in a perceptually annoying residual noise and speech distortion. In general, noise estimation is usually done by explicit detection of speech detection. This can be very difficult in the case of varying background noise. Furthermore, the background noise is assumed to be related stationary between speech pause. To overcome these problems, the noise spectrum needs to be estimated and updated continuously with a reliable speech detector. Among those algorithms, a recursive averaging is a commonly and easily used approach. Martin [1] proposed a method which is based on minimum statistic (MS). The noise spectrum estimation is obtained by tracking the minimum of the noisy speech power spectrum over a specific window. To improve the computational complexity of estimating noise spectrum, Doblinger [2] proposed an efficient method. However, it fails to differentiate between a rise in noise power and a rise in speech power. Further, Cohen et al. [3] introduced a MCRA approach to estimate noise power spectrum using a smoothing parameter which is defined as the speech-present probability in subbands. The speech-present probability in subbands of a given frame can be determined by the ratio between the noisy speech power spectrum to its local minimum over a period of 0.5-1.5 sec. Finally, the ratio is compared to a specific threshold value to decide updating noise power or not. In recently, Lin et al. [4] proposed a simple and reliable noise estimation technique. To estimate subband noise adaptively and continuously, the smoothing parameter is adjusted by a sigmoid function. However, a variable-level of noise is not considered in this case. We summarize that the drawback of most methods are slow in adapting to suddenly increase level of noise. In this paper, a noise estimator with rapid adaptation in a variable level noisy environment is presented. It depends only on the variation of the spectral energy but not on the amount of that. Based on the VAD, a noise estimator updates the noise spectrum fast and accurately even in …

Content truncated to 500 words