A series of carbonate-based lean-burn NOx trap (LNT) catalysts Pt−K 2 CO 3 /ZrO 2 with different K 2 CO 3 loading were prepared by sequential impregnation, which show extremely good performance for lean NOx storage and reduction. The catalyst containing 15 wt % K 2 CO 3 exhibits a large NOx storage capacity of 2.16 mmol/g and a very high NOx reduction percentage of 99%. Multiple techniques including X-ray diffraction (XRD), high-resolution transmission electron microscopy (HR-TEM), temperature-programmed decomposition (TPD), Fourier-transform infrared spectroscopy (FT-IR), and in situ diffuse reflectance infrared Fourier-transform spectroscopy (DRIFTS) were employed for catalyst characterization. The results of XRD, FT-IR, and HR-TEM conformably show that, at room temperature, the K species exist as amorphous K 2 CO 3 ; while at NOx storage temperature (350 °C), three kinds of K species including −OK groups, K 2 O, and K 2 CO 3 are simultaneously present in the catalysts as revealed by in situ DRIFTS, TPD, and FT-IR results. Surface carbonates are identified as the most active species for NOx storage, showing the best NOx storage performance. Higher K 2 CO 3 loading than 15 wt % leads to the formation of more bulk or bulk-like K 2 CO 3 species, which are unfavorable to NOx storage. As K 2 CO 3 loading is 10 wt % or less, the NOx is mainly stored as nitrates species such as monodentate nitrates, ionic nitrates, and bridging bidentate nitrates, while at higher K 2 CO 3 loading, the NOx is only stored as bidentate nitrite species. The presence of excess amount of K 2 CO 3 can decrease the ability of the catalysts for NO adsorption and oxidation, making the NOx oxidized only to nitrite species.
The performance of deep learning approaches to speech enhancement degrades significantly in face of mismatch between training and testing. In this paper, we propose a domain adversarial training technique for unsupervised domain transfer, that 1) overcomes domain mismatch, and 2) provides a solution to the scenario where we only have noisy speech data, and we don't have clean-noisy parallel data in the new domain. Specifically, our method includes two parts that are jointly trained, 1) an enhancement net to map noisy speech to clean speech by indirectly estimating a mask with a spectrum approximation loss, and 2) a domain predictor to distinguish between domains. As the proposed approach is able to adapt to a new domain only with noisy speech data in target domain, we call it an unsupervised learning technique. Experiments suggest that our approach delivers voice quality comparable with other supervised learning techniques that require clean-noisy parallel data.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.