21793 research outputs found
Sort by
On Zero-Shot Multi-Speaker Text-to-Speech Using Deep Learning
Get PDFThis thesis explores various aspects of zero-shot multi-speaker text-to-speech (TTS) synthesis using deep learning to create an effective system. A deep learning model for zero-shot multi-speaker TTS uses text and speaker identity as input to generate the respective output speech without fine-tuning for speakers not seen during training. The experiments consider a system with three main components: a speaker encoder network, a mel-spectrogram prediction network, and a vocoder network. A speaker encoder network captures the speaker identity in a fixed-sized speaker embedding. This speaker embedding is injected into a mel-spectrogram prediction network at one or more locations to generate a mel-spectrogram conditioned on the text and the speaker embedding. Finally, a vocoder network converts the mel-spectrogram into a waveform. All three components are trained separately. The speech synthesis aspects explored in the experiments include the speaker embedding injection method, speaker encoder network, speaker embedding injection location, and mel-spectrogram prediction network for the TTS system. The FiLM method from the visual reasoning field is adapted for the first time to inject speaker embeddings into the TTS workflow and compared against traditional methods. The significance of speaker embeddings is highlighted by comparing two well-established speaker embedding models. New combinations of speaker embedding injection locations are explored for two mel-spectrogram prediction networks. The best-performing model generates speech with naturalness ranging from fair to good, exhibits more than moderate speaker similarity, and shows potential for improvement. Additionally, the zero-shot multi-speaker TTS system is enhanced to generate fictitious voices
3D Point Cloud Reconstruction of a Single Image using Deep Learning
Get PDFAn essential yet challenging step in the 3D reconstruction problem is to train a machine or a robot to model 3D objects. Many 3D reconstruction applications depend on real-time data processing, so computational efficiency is a fundamental requirement in such systems. Despite considerable progress in 3D reconstruction techniques in recent years, developing efficient algorithms for real-time implementation remains an open problem. The present study addresses current issues in the high-precision reconstruction of objects displayed in a single-view image with sufficiently high accuracy and computational efficiency. To this end, we propose two neural frameworks: a CNN-based autoencoder architecture called Fast-Image2Point (FI2P) and a transformer-based network called Transformer+CNN (TransCNN3D). These frameworks consist of two stages: perception and construction. The perception stage addresses the understanding and extraction process of the underlying contexts and features of the image. The construction stage, on the other hand, is responsible for recovering the 3D geometry of an object by using the knowledge and contexts extracted in the perception stage. More specifically, the FI2P is a simple yet powerful architecture to reconstruct 3D objects from images faster (in real-time) without losing accuracy. Then, the TransCNN3D framework provides a more accurate 3D reconstruction without losing computational efficiency. The point cloud representation is used for the output of the reconstruction frameworks. The ShapeNet dataset is utilized to compare the proposed method with the existing ones in terms of computation time and accuracy. Simulations demonstrate the superior performance of the strategy
Human Sensorial Exploration in Designing a Comfortable Patient Room
Get PDFSome researchers have addressed what home means in architectural terms to human beings, and some others have investigated comfort and homeyness in relation to patients. Also, many studies, and on top of them Pallasmaa’s theory about the architecture of senses, have shown the importance of considering all human five senses in the design of a built environment and its positive effect on people’s sense of well-being. However, there has been no scientific evidence supporting the relationship between human senses and designing homely healthcare spaces which can result in the development of healing environments. So, this research investigates the comfort and homely design factors for patient rooms and their relationship to the five human senses in order to create multisensorial experiences. Also, this project aims to raise awareness among healthcare designers, caregivers, and patients in terms of the importance and role of considering all human senses in order to create comfortable and homely atmospheres in hospital rooms.
The current study is analytical and theoretical using qualitative research. Discourse analysis is conducted to find and evaluate the design recommendations based on sensorial qualities. Through the research-creation process, the current images of a cancer ward room in Montreal are analyzed. The final creation of this project is a webpage to reveal sensorial design recommendations as an information mechanism. The research-creation project has a didactic approach and tries to teach different design recommendations for a patient room and their sensorial aspects by interacting with the analyzed hospital room images using a website
Unraveling Post-Conflict Economic Recovery: Identifying Key Drivers of Above-Average Growth - A Cross-Country Analysis
Get PDFThis paper examines the factors influencing economic recovery in post-conflict settings, with a primary
emphasis on understanding the elements that contribute to heterogeneity in growth in postconflict
countries and identifying the specific variables associated with achieving above-average
growth rates. Utilizing panel data encompassing economic, social, and political variables, as well
as conflict-related data from 41 countries between 1970 and 2018, the study finds that terms of
trade, foreign direct investment (FDI), capital formation, and education play a crucial role in
influencing economic growth. These variables have a statistically significant impact on achieving
above-average growth rates in the aftermath of conflicts. However, the paper finds that constraints
on executives, serving as a proxy for institutions, turned out to be insignificant in relation
to post-conflict economic performance. Regarding the timing of recovery, the paper identifies an
initial 7-year post-conflict transition period marked by a notable catch-up effect. These findings
contribute valuable insights to the existing literature on post-conflict economic growth, enriching
our understanding of the dynamics of recovery in these challenging contexts and offering potential
implications for policies and interventions aimed at fostering sustainable progress and resilience
Development of a Secure Lossless Model for Image Steganography and its Hardware Realization for Real-Time Applications
Get PDFImage steganography is a promising security technology through which secret data is securely embedded in an image and then retrieved at the receiver upon receipt of the resulting image. The secret data is concealed in the cover image in a way such that it is invisible, and the resulting stego image is transmitted to an authorized receiver. The receiver has a key that enables extraction of the secret data from the stego image through a technique called recovery process. The design of a good image steganographic scheme aims at achieving three desirable characteristics: high embedding rate, imperceptibility of the secret data in the stego image, and robustness against attacks. Imperceptibility refers to the ability to maintain the quality of the cover image in the generated stego image after embedding the secret data. Embedding rate refers to the amount of secret data that is embedded in a single pixel of the stego image. Robustness against attacks refers to the ability to resist noise that could affect the stego image or steganalysis, ensuring that unauthorized receivers are not able to recover secret data. Existing image steganographic schemes lack providing both high embedding rate and high-quality stego image simultaneously as these are two mutually conflicting requirements. Moreover, robustness against attacks in these schemes is achieved using a high-complexity cryptographic algorithm that increases the time complexity of the image steganography scheme.
In this thesis, our objective is to design and implement a secure and lossless model for image steganography that has the three desirable features mentioned above. To achieve this objective, in the first part of the thesis (Chapters 3 and 4), we develop two novel modulus-based image steganographic algorithms that combine the benefits of two fundamental modulus-based schemes, namely, EMD and DE, resulting in the embedding rate and the quality of the stego image that are superior to that of the individual schemes. In the first algorithm proposed, the two embedding schemes are combined to embed a block of the secret data, whereas in the second algorithm proposed, each block of the secret data is first divided into two parts and then the first part is concealed by using one scheme and the second part by using the other scheme. To enhance the robustness of the proposed algorithms, the pixels of the cover image are first permuted using a low-complexity permutation mechanism prior to the concealment of the secret data. In the second part of the thesis (Chapters 5 and 6), hardware realizations of the two steganographic schemes developed in the first part of the thesis are provided using a reconfigurable platform.
In the third part of the thesis (Chapter 7), a secure lossless image steganographic model, consisting of the modules for encryption, embedding, and recovery, is presented, and a scheme for its efficient hardware implementation is proposed. To this end, first efficient hardware realizations for the encryption and recovery modules are developed. In the encryption module, secret data is made more secure by encrypting it using the hardware implementation of the private-key advanced encryption standard cryptosystem. In the recovery module, the quality of the cover image is preserved by encrypting the modified pixels and their original values using a hardware implementation of the public-key elliptic curve and Paillier cryptosystems. The hardware realizations of these two modules and that of the embedding module, designed in the second part of the thesis, are finally integrated together on an AMD Xilinx Zynq-7000 all programmable SoC platform. The efficiency and usefulness of the hardware implementation of the proposed steganographic model is demonstrated by considering the task of reducing the storage requirement for image data
The Hydrogeography of Mourning: Mapping the Life and Loss of Alberta Wetlands
Get PDFIn Alberta, 60-70 percent of wetlands have disappeared (Alberta Government, 2013, September 1). While this figure is used to quantify wetland loss in the province, it does not reflect the experiences of those emotionally impacted by such loss. Using Davidson and Milligan’s concept of emotional geography (2004) this thesis will explore grief in relation to wetlands within Alberta’s North Saskatchewan Watershed through the accounts and observations of some who inhabit the watershed and have been impacted by its transformation. Mapping the life and loss of wetlands is not a matter of locating geographical markers, rather it uses grief as a point of departure by making present the material, sensory, and emotional entanglements with wetlands, which then open to deeper research and analysis about wetland loss as part of Alberta’s settler history and ongoing economic development. For many who grew up on the Prairies, or who have spent a considerable amount of time with wetlands, grief is not only a response to their material loss, but rather a response to a disruption of one’s sense of being and place. Through the use of landscape ethnography and phenomenology as both a methodological and theoretical approach I examine the ways in which wetlands are not just backdrops to past experiences, but become part of living memory shaped in relation to kinship, home, and cultural politics. The North Saskatchewan Watershed is therefore a conceptual frame for imagining an emotional hydrogeography, one where wetland loss exposes a certain vulnerability in Being-with-wetlands, and in Being-without them
How Has Booktok Affected the Publishing Industry and Helped Self-Published Writers? A Study on Parasocial Relationships and Social Media Affordances.
Get PDFAlong with the many changes the pandemic brought, Tiktok grew in popularity, reaching new levels of users and downloads. As a result of the additional free time individuals had, many decided to pick up old hobbies or get into new ones, such as reading. The creation of Booktok, a subculture within the Tiktok application has brought on multiple changes and new opportunities within the publishing industry. This qualitative study conducted through interviews and netnography helps explore the effects of Booktok on the publishing industry. My findings show that Booktok contains mainly three actors: the supporter, the creator and the writer. Together these actors have created a community which has allowed for entertainment, parasocial relationships and professional relationships. While literature shows that parasocial relationships are evolving and changing from what they used to be, my findings add another perspective by demonstrating how Tiktok has taken these relationships to another level. The features offered on the digital platform allows for back-and-forth conversation between viewers and performers, which enhances the strength of these relationships. Additionally, this study looks at how social media affordances affect parasocial relationships. Writers, supporters and creators can use these new affordances, such as artistic extension, association, reenactment, engagement and support, promotional and educational tactics to promote and engage within the book community.
This allows for additional insight to the literature in terms of the use of social media. For practitioners the study emphasizes the changes within the publishing industry that publishers should be aware of and the new marketing tools that can be used to succeed as a writer.
Keywords: Tiktok, parasocial relationships, social affordances, booktok, publishing industry, self-publishing, booktoke
CEO Overconfidence and its Relationship with Overinvestment in the Context of the COVID-19 Pandemic
Get PDFThis paper primarily focuses on examining the correlation between CEO overconfidence and firm investment. Initially, I aimed to investigate whether there are any differences in confidence levels between male and female CEOs. The outcome of my research was that I was unable to identify any notable differences in the effects of CEO overconfidence on investment between male and female executives. This implies that regardless of gender, when a CEO displays overconfidence, they are likely to exhibit a similar inclination towards overinvestment. Additionally, my research focuses on the impact of the COVID-19 pandemic on CEO overconfidence and its subsequent influence on firm investment behavior. My study successfully establishes this connection. Specifically, in the current context of the COVID-19 pandemic, the relationship between CEO overconfidence and investment ratios is expected to weaken. However, the primary reason for this weakening effect is the volatile economic climate brought about by the COVID-19 era. The pandemic has significantly influenced companies to reduce their investments, resulting in the emergence of underinvestment as a prevalent issue, irrespective of the level of CEO confidence. In addition, my research reveals that a high level of CEO confidence is not able to substantially augment investment levels or adequately mitigate the problem of underinvestment in light of the substantial disruptions caused by the pandemic
Lignin-derived catalyst for biodiesel synthesis
Get PDFIncreasing dependence and overconsumption of non-renewable, pollutant-emitting fossil fuels has been declared as a primary contributor to climate change, prompting the shift to renewable energy sources. Biodiesel has emerged as a promising, more sustainable alternative energy source, primarily formed through either the transesterification of triglycerides or the esterification of free fatty acids into fatty acid methyl esters (biodiesel). However, current biodiesel production uses refined vegetable oils as feedstock which has raised concerns due to the competition with food crops and contributes a staggering 90% to the overall cost of production. Additionally, a major hurdle in the sustainability of biodiesel production is the current basic homogeneous catalysts employed react with free fatty acids in oils to produce soap, significantly contaminating the product and requiring extensive purification steps, resulting in substantial wastewater. For this reason, heterogeneous catalysts have garnered interest as more suitable alternatives. As well as being able to perform the transesterification of triglycerides, acidic catalysts also catalyse the esterification of free fatty acids present in most non-edible and waste cooking oils into biodiesel. In this study, a lignin-derived heterogeneous catalyst converted over 97% of oleic acid - a representative of free fatty acids - into biodiesel under optimal conditions. A basic heterogeneous catalyst was also synthesised for the transesterification of canola oil - a representation of triglycerides - into biodiesel under optimal conditions. The combination of these catalysts can be integrated in a two-step conversion of waste-cooking oil into biodiesel. Lignin is currently considered a waste material, being a by-product of the paper/pulp industry, with 98% being burnt for energy. The upcycling of lignin into a value-added product to convert waste cooking oil offers increased sustainability as well as access to environments and economies that are limited in resources and financials. Overall, the replacement of the standard basic homogeneous catalysts with solid heterogeneous ones overcomes the difficulty of separation and the employment of acidic catalysts allows utilisation of non-edible and waste cooking oils, eliminating majority of production costs and competition with the agriculture market
The Scattered Body of Truth: Contradictions, Ambiguities, and Transgressions in the Narrative of Paradise Lost
Get PDFScholars have expressed interest in ambiguous or discordant elements in John Milton’s Paradise Lost, but little attention has been devoted to the presence of these elements at the narrative level. Moreover, the attention previously given has largely been directed at resolving any apparent transgressions and showing the epic to be a unified didactic whole. This study aims to fill this gap by exploring the complexities of the narrative structure of Paradise Lost. I argue the transgressive narrative elements in the epic, which persistently resist understanding, are strategically employed and/or allowed in order to foster an acceptance of truth as inherently fragmented. I begin by outlining Milton’s definition of truth as presented in Areopagitica, which is used as a framework to understand the narrative ambiguities and contradictions. These transgressive components are subsequently analyzed across three dimensions of narrative: plot, voice, and temporality. Drawing on concepts from unnatural narrative theory, these analyses evidence how the transgressive features in the epic’s narrative often disrupt poetic stability, defy resolution, and deny readers a stable point of reference. Competing narratives of the same events, for example, present different and incompatible versions of the same event, and the reader does not know which, if any, is the accurate version. I conclude by considering the potential interpretive or reading strategies available to aid readers in maneuvering the narrative contradictions