Search

Extraction of relevant lip features is of continuing interest in the visual speech domain. Using end-to-end feature extraction can produce good results, but at the cost of the results being difficult for humans to comprehend and relate to. We present a new, lightweight feature extraction approach, motivated by human-centric glimpse-based psychological research into facial barcodes, and demonstrate that these simple, easy to extract 3D geometric features (produced using Gabor-based image patches), can successfully be used for speech recognition with LSTM-based machine learning. This approach can successfully extract low dimensionality lip parameters with a minimum of processing. One key difference between using these Gabor-based features and using other features such as traditional DCT, or the current fashion for CNN features is that these are human-centric features that can be visualised and analysed by humans. This means that it is easier to explain and visualise the results. They can also be used for reliable speech recognition, as demonstrated using the Grid corpus. Results for overlapping speakers using our lightweight system gave a recognition rate of over 82%, which compares well to less explainable features in the literature.

Biochar is considered a promising material for sequestering CO2 from the atmosphere, thus helping to alleviate climate change when returned to the soil. Biochar stability is the most decisive factor determining its C sequestration potential. Mineral modification may improve biochar characteristics, but systematic research on the effect of mineral modification on the C retention and stability of biochar and the associated mechanisms is limited. Therefore, in this study, rice straw was used to produce biochar at various temperatures (300, 400, 500, 600, and 700 °C), with vermiculite as a modified mineral material. Several methods including thermogravimetric analysis, Fourier transform infrared (FTIR) spectroscopy, X-ray photoelectron spectroscopy (XPS), and nuclear magnetic resonance (NMR) spectroscopy were used to evaluate the effect of vermiculite modification and carbonization temperature on biochar stability. Biochar yield and C retention ratio decreased with increasing temperatures but increased by 13.5–38.8% and 5.2–22.1%, respectively, after vermiculite modification. The ratios of C thermal weight loss, atomic H/C, and C oxidation loss in the biochar were reduced with increasing carbonization temperature, indicating improved thermal, aromatization, and chemical oxidation stability. A trade-off that did not compromise C sequestration potential was optimized at 700 and 600 °C for the unmodified and modified biochar, respectively. Furthermore, the total mineral content of the biochar, particularly Fe, Al, Mg, and Si, were increased by vermiculite modification. FTIR results showed that chemical bonds, such as Si–O–C and Fe–O, were formed or enhanced on the biochar surface after vermiculite modification. This was further certified by the XPS survey spectra. NMR results indicated that biochar stability was enhanced by increasing the aromatization rate during carbonization, that is, by the conversion of C from alkyl and carbonyl C to aromatic C. This study provides a basis for research into and the development of functional biochar and its application in C sequestration.

The extraction of relevant lip features is of continuing interest in the speech domain. Using end-to-end feature extraction can produce good results, but at the cost of the results being difficult for humans to comprehend and relate to. We present a new, lightweight feature extraction approach, motivated by glimpse based psychological research into racial barcodes. This allows for 3D geometric features to be produced using Gabor based image patches. This new approach can successfully extract lip features with a minimum of processing, with parameters that can be quickly adapted and used for detailed analysis, and with preliminary results showing successful feature extraction from a range of different speakers. These features can be generated online without the need for trained models, and are also robust and can recover from errors, making them suitable for real world speech analysis.