Now showing 1 - 5 of 67
- ItemMultilingual and Unsupervised Subword Modeling for Zero-Resource Languages(Elsevier Ltd, 2021-04) Hermann, Enno; Kamper, Herman; Goldwater, SharonSubword modeling for zero-resource languages aims to learn low-level representations of speech audio without using transcriptions or other resources from the target language (such as text corpora or pronunciation dictionaries). A good representation should capture phonetic content and abstract away from other types of variability, such as speaker differences and channel noise. Previous work in this area has primarily focused unsupervised learning from target language data only, and has been evaluated only intrinsically. Here we directly compare multiple methods, including some that use only target language speech data and some that use transcribed speech from other (non-target) languages, and we evaluate using two intrinsic measures as well as on a downstream unsupervised word segmentation and clustering task. We find that combining two existing target-language-only methods yields better features than either method alone. Nevertheless, even better results are obtained by extracting target language bottleneck features using a model trained on other languages. Cross-lingual training using just one other language is enough to provide this benefit, but multilingual training helps even more. In addition to these results, which hold across both intrinsic measures and the extrinsic task, we discuss the qualitative differences between the different types of learned features.
- ItemClassification learning of latent bruise damage to apples using shortwave infrared hyperspectral imaging(MDPI, 2021-07-22) Nturambirwe, Jean Frederic Isingizwe; Perold, Willem Jacobus; Opara, Umezuruike LinusBruise damage is a very commonly occurring defect in apple fruit which facilitates disease occurrence and spread, leads to fruit deterioration and can greatly contribute to postharvest loss. The detection of bruises at their earliest stage of development can be advantageous for screening purposes. An experiment to induce soft bruises in Golden Delicious apples was conducted by applying impact energy at different levels, which allowed to investigate the detectability of bruises at their latent stage. The existence of bruises that were rather invisible to the naked eye and to a digital camera was proven by reconstruction of hyperspectral images of bruised apples, based on effective wavelengths and data dimensionality reduced hyperspectrograms. Machine learning classifiers, namely ensemble subspace discriminant (ESD), k-nearest neighbors (KNN), support vector machine (SVM) and linear discriminant analysis (LDA) were used to build models for detecting bruises at their latent stage, to study the influence of time after bruise occurrence on detection performance and to model quantitative aspects of bruises (severity), spanning from latent to visible bruises. Over all classifiers, detection models had a higher performance than quantitative ones. Given its highest speed in prediction and high classification performance, SVM was rated most recommendable for detection tasks. However, ESD models had the highest classification accuracy in quantitative (>85%) models and were found to be relatively better suited for such a multiple category classification problem than the rest.
- ItemCanary in the coliform mine : exploring the industrial application limits of a microbial respiration alarm system(Public Library of Science, 2021-03-04) Stone, Wendy; Louw, Tobi M.; Booysen, Marthinus J.; Wolfaardt, Gideon M.; Zhang, DaweiFundamental ecological principles of ecosystem-level respiration are extensively applied in greenhouse gas and elemental cycle studies. A laboratory system termed CEMS (Carbon Dioxide Evolution Measurement System), developed to explore microbial biofilm growth and metabolic responses, was evaluated as an early-warning system for microbial disturbances in industrial settings: in (a) potable water system contamination, and (b) bioreactor inhibition. Respiration was detected as CO₂ production, rather than O₂ consumption, including aerobic and anaerobic metabolism. Design, thresholds, and benefits of the remote CO₂ monitoring technology were described. Headspace CO₂ correlated with contamination levels, as well as chemical (R² > 0.83–0.96) and microbiological water quality indicators (R² > 0.78–0.88). Detection thresholds were limiting factors in monitoring drinking water to national and inter- national standards (0 CFU/100 mL fecal coliforms) in both open- (>1500 CFU/mL) and closed-loop CO₂ measuring regimes (>100 CFU/100 mL). However, closed-loop detection thresholds allow for the detection of significant contamination events, and monitoring less stringent systems such as irrigation water (<100 CFU/mL). Whole-system respiration was effectively harnessed as an early-warning system in bioreactor performance monitoring. Models were used to deconvolute biological CO₂ fluctuations from chemical CO₂ dynamics, to optimize this real-time, sustainable, low-waste technology, facilitating timeous responses to biological disturbances in bioreactors.
- ItemWhich strategy saves the most energy for stratified water heaters?(MDPI, 2021) Ritchie, Michael J.; Engelbrecht, Jacobus A. A.; Booysen, M. J. (Thinus)ENGLISH ABSTRACT: The operation of water heating uses a substantial amount of energy and is responsible for 30% of a household’s overall electricity consumption. Determining methods of reducing energy demand is crucial for countries such as South Africa, where energy supply is almost exclusively electrical, 88% of it is generated by coal, and energy deficits cause frequent blackouts. Decreasing the energy consumption of tanked water heaters can be achieved by reducing the standing losses and thermal energy of the hot water used. In this paper, we evaluate various energy-saving strategies that have commonly been used and determine which strategy is best. These strategies include optimising the heating schedule, lowering the set-point temperature, reducing the volume of hot water used, and installing additional thermal insulation. The results show that the best strategy was providing optimal control of the heating element, and savings of 16.3% were achieved. This study also determined that the magnitude of energy savings is heavily dependent on a household’s water usage intensity and seasonality.
- ItemValuation of pumped storage in capacity expansion planning: a South African case study(MDPI, 2021) Van Dongen, Caroline; Bekker, Bernard; Dalton, AmarisENGLISH ABSTRACT: According to South Africa’s national energy policy, network penetration of variable renewable energy (VRE) generation will significantly increase by 2030. Increased associated network uncertainty creates the need for an additional flexible generation. As the planned VRE is mostly nonsynchronous PV and wind generators, additional ancillary services will also be required. Pumped Storage (PS), which is a well-established flexible generation technology with fast ramping capability and the ability to contribute various ancillary services, could help integrate increased VRE penetration on the South African network. However, in the latest revision of South Africa’s energy policy, PS was left out in favor of gas turbines and batteries as favored flexible generation options. This paper explores the two-part hypothesis that PS was disadvantaged in the formulation of a national energy mix due to: (a) ancillary services provided by PS not being explicitly monetized in energy modeling software; (b) the uncertainties associated with project costing assumptions. The value of PS in terms of providing ancillary services is firstly explored using the international literature. Secondly, the impact of input-cost uncertainties is demonstrated by comparing pumped storage, gas turbines, and batteries using levelized cost of energy (LCOE) curves and the Tools for Energy Model Optimization and Analysis (Temoa), North Carolina State University, USA, optimization software. Based on LCOE calculations using revised cost assumptions, it is found that PS may indeed be preferential to gas turbines or batteries, particularly at large load factors. The authors hope that this research contributes to the scientific understanding of the role that PS can play in supporting the integration of generation from renewable sources for effective grid operations.