Doctoral Degrees (Electrical and Electronic Engineering)

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 5 of 209
  • Item
    Multilingual acoustic word embeddings for zero-resource languages
    (Stellenbosch : Stellenbosch University, 2023-12) Jacobs, Chrsitiaan; Kamper, Herman; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Developing speech applications with neural networks require large amounts of transcribed speech data. The scarcity of labelled speech data therefore restricts the development of speech applications to only a few well-resourced languages. To address this problem, researchers are taking steps towards developing speech models for languages where no labelled data is available. In this zero-resource setting, researchers are developing methods that aim to learn meaningful linguistic structures from unlabelled speech alone. Many zero-resource speech applications require speech segments of different durations to be compared. Acoustic word embeddings (AWEs) are fixed-dimensional representations of variable-duration speech segments. Proximity in vector space should indicate similarity between the original acoustic segments. This allows fast and easy comparison between spoken words. To produce AWEs for a zero-resource language, one approach is to use unlabelled data from the target language. Another approach is to exploit the benefits of supervised learning by training a single multilingual AWE model on data from multiple well-resourced languages, and then applying the resulting model to an unseen target language. Previous studies have shown that the supervised multilingual transfer approach outperforms the unsupervised monolingual approach. However, the multilingual approach is still far from reaching the performance of supervised AWE approaches that are trained on the target language itself. In this thesis, we make five specific contributions to the development of AWE models and their downstream application. First, we introduce a novel AWE model called the Contrastive RNN. We compare this model against state-of-the-art AWE models. On a word discrimination task, we show that the Contrastive RNN outperforms all existing models in the unsupervised monolingual setting with an absolute Improvement in average precision ranging from 3.3% to 17.8% across six evaluation languages. In the multilingual transfer setting, the Contrastive RNN performs on par with existing models. As our second contribution, we propose a new adaptation strategy. After a multilingual model is trained, instead of directly applying it to a target language, we first _ne-tune the model using unlabelled data from the target language. The Contrastive RNN, although performing on par with multilingual variants, showed the highest increase after adaptation, giving an improvement of roughly 5% in average precision on five of the six evaluation languages. As our third contribution, we take a step back and question the effect a particular set of training languages have on a target language. We specifically investigate the impact of training a multilingual model on languages that belong to the same language family as the target language. We perform multiple experiments on African languages which show the benefit of using related languages over unrelated languages. For example, a multilingual model trained on one-tenth of the data from a related language outperforms a model trained on all the available training data from unrelated languages. As our fourth contribution, we showcase the applicability of AWEs by applying them to a real downstream task: we develop an AWE-based keyword spotting system (KWS) for hate speech detection in radio broadcasts. We validate performance using actual Swahili radio audio extracted from radio stations in Kenya, a country in Sub-Saharan Africa. In developmental experiments, our system falls short of a speech recognition based KWS system using five minutes of annotated target data. However, when applying the system to real in-the-wild radio broadcasts, our AWE-based system (requiring less than a minute of template audio) proves to be more robust, nearly matching the performance of a 30-hour speech recognition model. In the fifth and final contribution, we introduce three novel semantic AWE models. The goal here is that the resulting embeddings should not only be similar for words from the same type but also for words sharing contextual meaning, similar to how textual word embeddings are grouped together based on semantic relatedness. For instance, spoken instances of \football" and \soccer", although acoustically different, should have similar acoustic embeddings. We specifically propose leveraging a pre-trained multilingual AWE model to assist semantic modelling. Our best approach involves clustering word segments using a multilingual AWE model, deriving soft pseudo-word labels from the cluster centroids, and then training a classifier model on the soft vectors. In an intrinsic word similarity task measuring semantics, this multilingual transfer approach outperforms all previous semantic AWE methods. We also show for the first time that AWEs can be used for downstream semantic query-by-example search.
  • Item
    Engineering carbon nanotubes as therapeutic nanocarriers of Tulbaghia violacea, Annona muricata, Dicoma capensis and Dodonaea viscosa plant-based extracts, targeting breast and colorectal cancer
    (Stellenbosch : Stellenbosch University, 2023-12) Gwanzura, Takunda; Perold, Willem; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Cancer is one of the most prevalent diseases globally and it is characterized by uncontrolled rapid cell division and differentiation. Lack of tumour specificity, dose-related toxicity and low bioavailability of chemotherapy drugs are major hindrances to cancer treatment. Nanotechnology has given the platform to selectively interact with cancerous cells and increase cellular uptake and drug localization. Functionalisation of nanoparticles can be done to recognize cancer cells and giving accurate and selective drug delivery which does not interact with healthy cells. In order to develop more efficient therapeutic regimes, a better understanding of the type of nanoparticles suitable for drug delivery is required. Over the past years, carbon nanotubes have been used as nanocarriers to transport anticancer drugs, genes, and proteins for chemotherapy. Furthermore, the possibility of conjugating carbon nanotubes with anticancer plant-based drugs creates advanced therapeutic applications. Therefore, the aim of this project was to develop a single-walled carbon nanotube (SWCNT) nanocarrier bio-conjugated with plant-based bioactive compounds which can target can- cer cells specifically. The first phase of the study involved purification and functionalisation of carbon nanotubes. Hydrochloric acid was used to purify the carbon nanotubes and functionalisation was done with polyethylene glycol (PEG) and folic acid (FA). Fourier transform infrared (FTIR) spectroscopy was used to confirm functionalisation. Four plants were identified and used, namely Annona muricata, Dodonaea viscosa, Dicoma capensis and Tulbaghia violacea. Two plant extraction methods were assessed, and the closed loop extraction method obtained the most bioactive compounds from the plant extracts. The results were confirmed by liquid chromatography-mass spectrometry (LC-MS) analysis. This was followed by bio-conjugation of the functionalised carbon nanotubes with bioactive compounds from the four plants. Ultraviolet–visible (UV-Vis) spectroscopy was used to confirm bio-conjugation. In vitro cytotoxicity studies were undertaken to assess the effects of bioactive compounds and bio-conjugates in breast cancer cell lines (MCF-7 and MDA-MB-231), a colon cancer cell line (HT-29) and a non-tumorigenic breast epithelial cell line (MCF-12A). The in vitro cytotoxicity results showed a low cell viability for cancer cell lines whilst that for normal cells remained higher. The half-maximal inhibitory concentration (IC50) values were determined, and all plants showed a value lower than 30 μg/ml. The selectivity index (SI) of the plant extracts was also calculated and all plants showed a high SI value greater than 2. The bio-conjugates showed a higher cell viability in normal cells and a lower cell viability in cancer cells compared to plant extracts alone, which was due to the conjugation with carbon nanotubes improving selectivity and efficacy. Furthermore, the bio-conjugate with all four plant extracts mixed had the lowest cell viability in all cancer cell lines indicating synergism. The mechanisms of cellular uptake were determined by fluorescence microscopy and it was observed that folate receptor mediated endocytosis and caveolae mediated endocytosis both took place for bio-conjugate cellular internalization. The type of cell death occurring in cancer cells was determined by flow cytometry studies. It was confirmed that both apoptosis and necrosis took place in the cancer cells. In conclusion, the carbon nanotubes were successfully used as nanocarriers and their bio-conjugation with plant based bioactive compounds improved efficacy and selectivity towards cancer cells in this study.
  • Item
    Impact and cause of sensitivity ripple in radio astronomy reflector antennas
    (Stellenbosch : Stellenbosch University, 2023-12) Cerfonteyn, William; De Villiers, Dirk; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: This dissertation presents a study on the impact and cause of the frequency ripple in receiving sensitivity of electrically small reflector a ntennas. Knowledge of the shape and spectral content of the ripple is important in some radio astronomy applications. Although no alternative to high fidelity s ampling of the antenna response, using appropriate computational electromagnetic simulations, was found to accurately characterize the ripple response, the different physical causes of the ripple, and their relative impact on the final response, is comprehensively considered. For next-generation telescopes using wide-band room temperature low-noise amplifiers ( LNA), as opposed to extremely cold cryogenic systems, it is shown that the ripple may, in many cases, be reliably ignored during the initial design phase of the system - even for electrically very small systems. It is further illustrated how the ripple characteristics vary as a function of antenna pointing angle, and how, in some cases, the spillover energy onto the hot ground may dominate the effect. To date, such characterizations have been ignored in the literature, and focus has mainly been on the behaviour of the antenna main beam - which normally points at a relatively cold sky. The dissertation describes that the cause of the frequency ripple in receiving sensitivity is due to non-ideal effects. The sensitivity ripple is influenced only by the ripple in the antenna noise temperature (ANT), and the ripple in the aperture efficiency (AE), while the antenna and LNA ar e we ll ma tched. Furthermore, the ripple of the ANT and AE is determined only by the radiation intensity ripple, which is caused by stray radiation, due to non-ideal effects, interfering with the radiation pattern of the full reflector system. Non-ideal radiation or effects occur, when the reflector does not operate ideally, which occurs when the reflector is not infinitely large. The extent of the non-ideal radiation is correlated to the electrical size of the reflector, and thus electrically small reflectors start to diverge more from the ideal radiation proposed by geometric optics. Furthermore, it is highlighted that the ANT is a function of all directions, and thus the sensitivity is also. This results in certain directions or regions being significantly more important than others for the ANT calculation, in a specific pointing angle. In these regions, the ripple in the radiation pattern is observed in the ripple of the ANT, as expected. Heatmaps are constructed to illuminate these important angles which can be used to gain insight into which non-ideal effects dominate the ripple contribution, and prove the strong dependence of the ANT ripple on the pointing angle. Besides being a function of all directions, the ANT is also a function of many physical parameters. Some of these parameters and their effect on the ANT is investigated. During the design of radio telescope projects, such as the ngVLA, state-of-the art estimations for ANT and AE are used. The accuracy of these approximations for ANT and AE are investigated, and characterised. These strategies used for rapid approximation are fast, however, often neglect modeling the ripple. This is because precise calculation of the ripple is often expensive in terms of computation and storage, and usually not necessary during the optimisation phase. The modeling efficiency of these techniques is interrogated, which is a key component in the effective designing of reflectors for radio astronomy. Physical Optics (PO) simulation strategies are often used in larger radio telescope designs, compared to Method of Moments (MoM). For smaller designs, the accuracy between these techniques becomes important to consider. MoM accounts for more non-ideal effects, compared to PO, and as such models the ripple more accurately. In small designs, the Physical Theory of Diffraction (PTD) can be used in conjunction with PO, to more accurately model the influence of non-ideal effects. There is a breakpoint in frequency, where the ripple modeled with MoM and PO (with PTD) will converge, as the electric size of the reflector increases. These techniques are compared and analysed, to characterise their impact for use in modeling the ripple of the ANT in smaller designs. Finally in the conclusion, future work is considered, where possible ripple prediction methods are discussed. One of these methods uses a combination of techniques (including Validated Exponential Analysis or VEXPA) to recover a unique signal composition, from a sampling rate under the Nyquist rate. Besides this method, the viability of geometric arguments, or applying a preprocessed ripple, is considered for ripple prediction. The exclusive use of far-fields in the sensitivity calculation, without considering the near-field, is also discussed, with suggestions to aid the investigation of its effect.
  • Item
    Probabilistic conflict prediction: an accurate and computationally efficient approach
    (Stellenbosch : Stellenbosch University, 2023-12) Roelofse, Christiaan Roelofse; Van Daalen, Corne; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Collision (or conflict) prediction is a vital component of motion planning for autonomous vehicles to ensure safe operation, both in the context of autonomous navigation and in the context of an advisory system for manned vehicles. Prediction methods must be accurate to know whether motion planning corrections are required. However, computationally efficient prediction methods are Essential in order to ensure that motion planning corrections are brought about in a timely manner. Efficient prediction methods are especially crucial when testing large sets of candidate trajectories for conflict, given the accumulation of computational cost for each candidate. This dissertation presents a probabilistic conflict prediction method that demonstrates the same accuracy as existing methods, but at a significantly reduced computational cost. This is achieved by a novel reformulation of the conflict prediction problem in terms of the first-passage time using a dimension-reduction transform. First-passage time distributions are analytically derived for a subset of Gaussian motion models which describe vehicle motion. The proposed method is applicable for stochastic processes where the vehicle mean motion can be approximated by linear segments, and the conflict boundary is modelled as – or approximated by – either piece-wise straight lines in 2-D, or neighbouring planes in 3-D. The proposed method was tested in simulation and compared to state-of-the-art conflict prediction methods. These comparison methods consist of two probability flow methods, as well as an instantaneous conflict probability method. The results demonstrate a significant decrease of computation time.
  • Item
    The development of a biosensor for the early detection of pancreatic cancer
    (Stellenbosch : Stellenbosch University, 2023-03) Ebrahim, Taskeen; Perold, Willem; Engelbrecht, Anna-Mart; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Pancreatic cancer has one of the highest cancer mortality rates, as it is often detected in late stages. Current methods of detection include diagnostic imaging tests or laboratory intensive blood tests such as radioimmunoassays or ELISAs. Researchers have used biosensors for detection and monitoring of different medical conditions, as a more accessible and cheaper alternative. This project focuses on the design of a biosensor toward the early detection of pancreatic cancer, using Carbohydrate Antigen 19-9 (CA19-9) as the selected biomarker for the biosensor. Electrochemical impedance spectroscopy was identified as an appropriate transducer mechanism for this biosensor and uses gold interdigitated electrodes as a sensor transducer surface. Anti-CA19-9 antibodies were immobilized onto gold using covalent bonding and crosslinking chemistries, and binding was validated using fluorescence microscopy. After electrodes were electronically characterized to identify the appropriate impedance and frequency ranges, an impedance analyser was designed, fabricated, and tested, with added computation of complex capacitance. The impedance analyser was calibrated and tested relative to the PalmSens4 Electrochemical Interface. The designed impedance analyser showed mean impedance and phase errors of 0.538% and 0.381% respectively. Similarly, the accuracy of complex capacitance computation showed errors of 1.222% and 0.656% for real and imaginary components respectively. The impedance analyser differentiated between changes in concentration of phosphate buffered saline using complex capacitance. The biosensor was tested with five concentrations of CA19-9 and differentiated between concentrations above and below the thresholds for pancreatic cancer. The design of a novel impedance analyser utilizing complex capacitance computation for the quantification of CA19-9 using IDEs fabricated on an FR4 board is a unique contribution to biosensor research.