Search

Recent research advances in deep neural network (DNN)-based beamformers have shown great promise for speech enhancement under adverse acoustic conditions. Different network architectures and input features have been explored in estimating beamforming weights. In this paper, we propose a deep beamformer based on an efficient convolutional recurrent network (CRN) trained with a novel ARray RespOnse-aWare (ARROW) loss function. The ARROW loss exploits the array responses of the target and interferer by using the ground truth relative transfer functions (RTFs). The DNN-based beamforming system, trained with ARROW loss through supervised learning, is able to perform speech enhancement and speaker localization jointly. Experimental results have shown that the proposed deep beamformer, trained with the linearly weighted scale-invariant source-to-noise ratio (SI-SNR) and ARROW loss functions, achieves superior performance in speech enhancement and speaker localization compared to two baselines.

Abstract A three-stage approach is proposed for speaker counting and speech separation in noisy and reverberant environments. In the spatial feature extraction, a spatial coherence matrix (SCM) is computed using whitened relative transfer functions (wRTFs) across time frames. The global activity functions of each speaker are estimated from a simplex constructed using the eigenvectors of the SCM, while the local coherence functions are computed from the coherence between the wRTFs of a time-frequency bin and the global activity function-weighted RTF of the target speaker. In speaker counting, we use the eigenvalues of the SCM and the maximum similarity of the interframe global activity distributions between two speakers as the input features to the speaker counting network (SCnet). In speaker separation, a global and local activity-driven network (GLADnet) is used to extract each independent speaker signal, which is particularly useful for highly overlapping speech signals. Experimental results obtained from the real meeting recordings show that the proposed system achieves superior speaker counting and speaker separation performance compared to previous publications without the prior knowledge of the array configurations.

For precise delineation of glioma extent, amino acid PET is superior to conventional MR imaging. Since metabolic MR sequences such as chemical exchange saturation transfer (CEST) imaging and MR spectroscopy (MRS) were developed, we aimed to evaluate the diagnostic accuracy of combined CEST and MRS to predict glioma infiltration. Eighteen glioma patients of different tumor grades were enrolled in this study; 18F-fluoroethyltyrosine (FET)-PET, amide proton transfer CEST at 7 Tesla(T), MRS and conventional MR at 3T were conducted preoperatively. Multi modalities and their association were evaluated using Pearson correlation analysis patient-wise and voxel-wise. Both CEST (R = 0.736, p < 0.001) and MRS (R = 0.495, p = 0.037) correlated with FET-PET, while the correlation between CEST and MRS was weaker. In subgroup analysis, APT values were significantly higher in high grade glioma (3.923 ± 1.239) and IDH wildtype group (3.932 ± 1.264) than low grade glioma (3.317 ± 0.868, p < 0.001) or IDH mutant group (3.358 ± 0.847, p < 0.001). Using high FET uptake as the standard, the CEST/MRS combination (AUC, 95% CI: 0.910, 0.907–0.913) predicted tumor infiltration better than CEST (0.812, 0.808–0.815) or MRS (0.888, 0.885–0.891) alone, consistent with contrast-enhancing and T2-hyperintense areas. Probability maps of tumor presence constructed from the CEST/MRS combination were preliminarily verified by multi-region biopsies. The combination of 7T CEST/MRS might serve as a promising non-radioactive alternative to delineate glioma infiltration, thus reshaping the guidance for tumor resection and irradiation.