A Probability Weighted Beamformer for Noise Robust ASR

Abstract

We investigate a novel approach to spatial filtering that is adaptive to conditions at different time-frequency (TF) points for noise removal by taking advantage of speech sparsity. Our approach combines a noise reduction beamformer with a minimum variance distortionless response (MVDR) beamformer or Generalized Eigenvalue (GEV) beamformer through TF posterior probabilities of speech presence (PPSP). To estimate PPSP, we study both statistical model-based and neural network based methods, where in the former, we use complex Gaussian mixture modeling (CGMM) on temporally augmented spatial spectral features, and in the latter, we use neural network (NN) based TF masks to initialize speech and noise covariance matrices in CGMM. We have conducted experiments on CHiME-3 task. On its real noisy speech test set, our methods of feature augmentation, TF dependent spatial filter, and NN-based mask initialization on covariances for CGMM have yielded relative word error rate (WER) reductions cumulatively by 8%, 16%, and 25% over the original CGMM based MVDR. On the real test data, the three methods have also produced consistent WER reductions when replacing MVDR by GEV.

Multiple beamformers with ROVER for the CHiME-5 Challenge

“…We design 5 FBs focusing on 60, 90, 120, 150 and 180 degree respectively. Apart from 5 FBs, complex Gaussian mixture model based MVDR (CGMM-MVDR) beamforming, which has been proved to be effective in previous work [3,4], is also applied to provide the 6th speech stream.…”

A Novel Method to Correct Steering Vectors in MVDR Beamformer for Noise Robust ASR

“…Instead, [12] maximizes the beamformer output power under the constraint that the estimated SV does not converge to any interference direction. Recently, [13,4,14,15] used a time-frequency (TF) mask-based approach to beamforming without imposing a priori assumptions on SVs. The SVs are estimated solely from the complex Gaussian mixture model (CGMM) based masks and the observation data.…”