Single Channel Speech Enhancement Using Adaptive Soft-Thresholding with Bivariate EMD

Abstract

This paper presents a novel data adaptive thresholding approach to single channel speech enhancement. The noisy speech signal and fractional Gaussian noise (fGn) are combined to produce the complex signal. The fGn is generated using the noise variance roughly estimated from the noisy speech signal. Bivariate empirical mode decomposition (bEMD) is employed to decompose the complex signal into a finite number of complex-valued intrinsic mode functions (IMFs). The real and imaginary parts of the IMFs represent the IMFs of observed speech and fGn, respectively. Each IMF is divided into short time frames for local processing. The variance of IMF of fGn calculated within a frame is used as the reference term to classify corresponding noisy speech frame into noise and signal dominant frames. Only the noise dominant frames are soft-thresholded to reduce the noise effects. Then, all the frames as well as IMFs of speech are combined, yielding the enhanced speech signal. The experimental results show the improved performance of the proposed algorithm compared to the recently reported methods.

Suppression of Non-Stationary Noises Through the Generalized Signal Detector

“…Several approaches are developed in earlier based on EMD [6]. In [6], an adaptive soft thresholding algorithm was developed based on EMD.…”

“…In [6], an adaptive soft thresholding algorithm was developed based on EMD. Here the variance is derived for every IMF not for speech signal by which this algorithm achieved better performance.…”

“…Here the above eq. (6) (2) and (6). Deciding that the speech is not present ( 0 ) when there is a presence of speech ( 1 ) makes the speech enhancement not to achieve an optimal performance.…”

Empirical Mode Decomposition for Advanced Speech Signal Processing

“…The expression to calculate the optimum adaptation factor (λ opt ) is defined as (19) to fit the data points (v i , w i ), i=1,2,….., d(=9); where v i and w i are the input SNR and optimum value of λ (to obtain the maximum output SNR) respectively as listed in Table 2. We experimentally found that there is a nonlinear relation between the input SNR and adaptation factor, for that we choose a third degree polynomial to fit the non-linear data points with minimum stable coefficients [40]. To obtain the coefficients, Eq.…”

“…Such method degrades the performance of speech enhancement when the speechless is not detected perfectly. Bivariate EMD (bEMD) based approach is implemented for effective noise estimation to derive the parameters required for soft-thresholding [41]. In bEMD, two variables are decomposed simultaneously without losing mutual dependency.…”