Probabilistic integration of diffuse noise suppression and dereverberation

Abstract

This paper deals with joint suppression of diffuse noise and reverberation, to enhance perceived speech quality and speech recognition performance. Although diffuse noise and reverberation are both omnipresent in the real world, conventional methods have modeled only one while neglecting the other. In contrast, we propose a novel joint suppression method that employs a unified probabilistic model of observed signals affected by both diffuse noise and reverberation. Through likelihood maximization, this unified model enables proper parameter estimation that takes into account both diffuse noise and reverberation. As a byproduct, we also propose a novel method for diffuse noise suppression. Experimental results demonstrate the effectiveness of the proposed joint suppression method in terms of dereverberation and denoising.

Multichannel Identification and Nonnegative Equalization for Dereverberation and Noise Reduction Based on Convolutive Transfer Function

“…This zeromagnitude region is caused by the filter banks, and thence is also present in the frequency response of filter b p,k . This common zeros of two channels are problematic for the crossrelation method (7). It can be seen from Fig.…”

“…Multichannel dereverberation broadly includes the following techniques: (i) the spectral enhancement technique removes reverberation by spectral subtraction. Many techniques have been proposed to estimate the power spectral density of reverberation, such as statistical model [2], coherent-to-diffuse power ratio [3]; (ii) the linear prediction technique is used in the weighted prediction error (WPE) algorithm [4], [5], [6]: reverberation is first estimated by filtering the microphone signal with the prediction filter, and then subtracted from the microphone signal (in a way, the prediction filter could be considered as the inverse of the channel filter, which is estimated by maximizing a likelihood function derived from a timevarying all-pole model of source spectra); (iii) probabilistic techniques apply dereverberation by maximizing the likelihood of the generative model of the microphone signals, such as in [7], [8], and (iv) the multichannel equalization technique first blindly identifies the channel filters, then applies the inverse filtering on the microphone signals.…”

Multi-Channel Linear Prediction-Based Speech Dereverberation With Sparse Priors

“…The WPE method has been modified to include pre-trained log-spectral priors in [22], and a time-varying Laplacian model for the desired speech signal has been used in [36]. Recently, several methods based on auto-regressive modeling have been proposed, aiming to address noisy [37], [38] and time-varying acoustic scenarios [19] with multiple sources [18], [19], [39].…”

“…The logarithmic penalty is related to the -norm through [46]. Moreover, the set of local minima of the optimization problem in (36) corresponds to the set of local minima of the optimization problem [46] (37) Using (8) the desired speech signal can be further expressed as (38) with (39) where is equivalent to the prediction vector . Now the optimization problem (35) can be rewritten directly in terms of the prediction vector as (40) where…”

“…The LS optimization problem in (41) has a closed-form solution (42) that is equivalent to estimating the prediction vector in (14). The estimate of the desired signal in the -th iteration is given using (38) as . As in [29], [49], [52], the weights are updated in each iteration as (43) so that the cost function in (41) is a first-order approximation of the cost function in (40).…”