11723 research outputs found
Sort by
Entwicklung vielseitiger Haarfärbemittel auf Basis von Indigofera tinctoria L.: Ein Beitrag zur nachhaltigen Kosmetikchemie
Get PDFPlant hair dyes based on Indigofera tinctoria L. show great potential for the development of sustainable and effective hair dye formulations. The enzymatic hydrolysis of the dye precursor indican by β-D-glucosidase in an aqueous environment leads to the formation of indoxyl, which penetrates the hair fiber and oxidizes there to indigo and indirubin. Both dye molecules are insoluble in water and permanently anchored in the hair cortex.
The hair structure has a significant influence on the color expression, in particular the subsequent formation of indirubin. This leads from the initial blue tone to a violet color impression. With 5% (w/v) dried and ground indigo leaves in an aqueous solution, a clear color shift of ΔE = 33.0 is evident after 14 days, with permanent color stability over 30 hair washes. Furthermore, the addition of isatin and cysteine or ascorbic acid achieves an immediate and stable indigo-red coloration on the hair for the first time (ΔE day 0 and day 14: = 7.6), in accordance with EU cosmetics guidelines.
The combination of indican, β-D-glucosidase and water, essential for color formation from Indigofera tinctoria L., must be taken into account when preparing extracts for product development. A heat-denatured, aqueous indigo extract containing indican offers advantages for product development and enables controlled dye release through the targeted addition of β-D-glucosidase immediately before application. In addition, it shows particularly low color changes on the hair within 14 days (ΔE = 1.6).
Furthermore, important factors for the cultivation and processing of the plant in India can be identified in order to ensure dye quality.
Finally, based on the totality of investigations, a new color formation pathway of Indigofera tinctoria L. on the hair is presented, starting from the storage form in plants, indican. The results prove that functionality and sustainability are compatible in hair coloring with Indigofera tinctoria L.. The targeted control of color formation
enables the development of bio-based hair dyes with high performance and ecological compatibility - a promising alternative to conventional synthetic products
Coexisting Invariants, Dynamical Spin Control, and Exchangeless Braiding A Hitchhiker’s Guide to Topology
Get PDFDie moderne Theorie der kondensierten Materie greift zunehmend auf topologische Ideen zurück, um Phänomene zu erklären, die sich rein lokalen Beschreibungen entziehen.
Während die theoretischen Grundlagen der topologischen Quantenmaterie gut etabliert sind, kann noch viel darüber gelernt werden, wie topologische Eigenschaften mit anderen Aspekten physikalischer Systeme interagieren.
In dieser Dissertation wird das Zusammenspiel zwischen Topologie und konkurrierenden physikalischen Effekten anhand einer Reihe von quantenmechanischen und quantenklassischen Fallstudien untersucht. Diese umfassen Systeme mit koexistierenden topologischen Ordnungen, quantenklassische Dynamik und kleine Quanteninformationsplattformen.
Die erste Studie befasst sich mit den Auswirkungen koexistierender topologischer Ordnungen in einem quantenklassischen Hybridsystem, in dem klassische Störstellenspins an ein Haldane-Modell von Elektronen mit Spin-1/2 auf einem periodischen Honigwabengitter gekoppelt sind.
In diesem Rahmen stellt der Konfigurationsraum der klassischen Störstellenspins eine Parametermannigfaltigkeit dar, die eine extrinsische monopolartige Klassifizierung ermöglicht, welche die intrinsische Chern-Klassifizierung des Haldane Trägersystems ergänzt.
Die zusätzliche monopolartige topologische Ordnung erklärt das Auftreten eines spektralen Flusses von Energien gebundener Zustände, welcher die Energielücke des Haldane-Modells als Funktion der Austauschkopplungsstärke überbrücken.
Die Form dieses spektralen Flusses wird derweil durch die Chern-Topologie des Trägersystems bestimmt, die festlegt, ob im Grenzfall unendlich großer Austauschkopplungsstärken zwischen den Störstellenspins und dem Trägersystem gebundene In-Gap-Zustände auftreten oder nicht.
Die Koexistenz der konventionellen Chern Impulsraumtopologie und der monopolartigen Konfigurationsraumtopologie ermöglicht die numerische Konstruktion topologischer Phasendiagramme, die wertvolle Einblicke in die wechselseitigen Beziehungen zwischen den beiden topologischen Strukturen liefern.
Die zweite Studie zeigt, wie die topologischen Randzustände eines Quanten-Spin-Hall-Systems zur Steuerung der Echtzeitdynamik einer magnetischen Störstelle genutzt werden können.
Zu diesem Zweck wird ein klassischer Störstellenspin an eine Kante eines Kane-Mele-Modells auf einem endlichen Streifensegment gekoppelt.
Die Echtzeitentwicklung des resultierenden quantenklassischen Hybridsystems wird ermittelt, indem der vollständige Satz gekoppelter Bewegungsgleichungen des klassischen Spins und des elektronischen Systems numerisch gelöst wird.
Um eine Simulation der Langzeitdynamik zu ermöglichen, werden dissipative Randbedingungen für alle Ränder mit Ausnahme des Randes, der den Störstellenspin trägt, eingeführt.
Die feste Kopplung zwischen Spin- und Impulsrichtung in den Kane-Mele-Randmoden führt dazu, dass sich Spindichte-Injektionen unidirektional und nahezu verlustfrei ausbreiten, was eine gezielte Manipulation der Störstellenspindynamik aus der Ferne ermöglicht.
Iterierte Protokolle aus Spininjektion, unidirektionaler Ausbreitung und Streuung am Störstellenspin ermöglichen die Implementierung eines vollständigen und reversiblen Spinschaltprozesses.
Die dritte Studie nutzt Rotationen der supraleitenden Phase in kleinen Netzwerken aus schwach verknüpften Kitaev-Ketten, um logische Braidingoperationen für topologisches Quantencomputing zu realisieren, ohne die anyonischen Majorana-Quasiteilchen auszutauschen.
Die Braidingprotokolle werden numerisch verifiziert, indem die nicht-abelsche Wilczek-Zee-Phase des niederenergetischen Vielteilchen-Unterraums unter Verwendung der Bertsch-Robledo-Überlappformel ausgewertet wird.
Der Parameterraum von zwei schwach verknüpften Kitaev-Ketten teilt sich in zwei Regionen, die durch unterschiedliche Braidingresultate gekennzeichnet sind, nämlich projektive X- oder Z-artige Braidingoperationen.
Eine Auswahl repräsentativer Phasendiagramme zeigt, dass die X- und Z-Phasen durch einen kontinuierlichen Übergang voneinander getrennt sind.
Dieser kontinuierliche Übergang ist darüber hinaus auf einer scharfen Übergangs-Hyperfläche zentriert,
welche aus einem Minimalmodell der vier beteiligten Majorana-Moden abgeleitet wird.
Die Studie zeigt, dass die anyonischen Eigenschaften selbst bei endlichen Systemgrößen und schwacher Kopplung zwischen den Kitaev-Ketten erhalten bleiben.Modern condensed matter theory increasingly draws on topological ideas to explain phenomena that defy purely local descriptions.
While the theoretical foundations of topology in quantum matter are well established, much can still be learnt about how topological properties interact with other aspects of physical systems.
This thesis explores the interplay between topology and competing physical effects through a series of quantum and quantum-classical case studies.
We cover systems with coexisting topological orders, quantum-classical dynamics, and small-scale quantum information platforms.
The first study addresses the implications of coexisting topological orders in a quantum-classical hybrid system formed by classical impurity spins coupled to a spinful Haldane model on a periodic honeycomb lattice.
In this setting, the configuration space of the classical impurity spins constitutes a parameter manifold that enables an extrinsic monopole-like classification complementing the intrinsic Chern classification of the Haldane host.
The additional monopole-like topological order explains the emergence of a spectral flow of bound-state energies bridging the host's energy gap as a function of the exchange-coupling strength.
The form of this spectral flow, however, is determined by the Chern topology of the host system, which dictates whether or not impurity-bound in-gap states appear in the limit of infinitely large exchange coupling strengths between impurity spins and host.
The coexistence of the conventional Chern momentum-space topology and the monopole-like configuration-space topology enables the numerical construction of topological phase diagrams that provide valuable insights into the interrelations between the two topological structures.
The second study demonstrates how the topological edge states of a quantum spin Hall system can be harnessed to control the real-time dynamics of a magnetic impurity.
To this end, a classical impurity spin is coupled to one edge of a Kane-Mele model on a finite ribbon segment.
The real-time evolution of the resulting quantum-classical hybrid system is obtained by numerically solving the full set of coupled equations of motion for the classical spin and the electronic system.
In order to enable the simulation of long-time dynamics, dissipative boundary conditions are imposed on all but the impurity-hosting edge.
Spin density injections into the spin-momentum locked Kane-Mele edge modes propagate unidirectionally and with minimal loss, enabling remote manipulations of the impurity spin dynamics.
Iterated protocols of spin injection, unidirectional propagation, and scattering off the impurity spin allow the implementation of a complete and reversible spin switching process.
The third study employs superconducting phase rotations in small networks of weakly-linked Kitaev chains to realise logical braiding operations for topological quantum computation without exchanging anyonic Majorana quasiparticles.
Braiding protocols are verified by numerically evaluating the non-Abelian Wilczek--Zee phase of the low-energy many-body subspace using the Bertsch-Robledo overlap formula.
The parameter space of two weakly linked Kitaev chains divides into two regions characterised by distinct braiding outcomes, i.e. projective X- or Z-type braiding operations.
A selection of representative phase diagrams reveal that the X and Z phases are separated by a continuous crossover.
Moreover, this crossover regime is centred on a sharp transition hypersurface derived from a minimal model of the four involved Majorana modes.
The study demonstrates the robustness of anyonic properties against finite-size effects and weak couplings between Kitaev chains
Auswirkungen einer konditionellen Expression des Wachstumsfaktors ERBB2 auf künstliches Stammzell-abgeleitetes Herzgewebe (Engineered Heart Tissue, EHT)
Get PDFKardiovaskuläre Erkrankungen sind die häufigste Ursache für ein frühzeitiges Ableben.
Aktuelle Therapieoptionen können zwar unter bestimmten Bedingungen die Symptome nach
einem Herzinfarkt mit Verlust von Herzgewebe lindern, doch die Revitalisierung des
entstandenen Narbengewebes bleibt weiterhin eine Herausforderung. Experimentelle
Therapien wie die Transplantation von humanen induzierten pluripotenten Stammzellen
(hiPSC) abgeleiteten Kardiomyozyten können prinzipiell abgestorbenes Myokardgewebe
durch neue Kardiomyozyten ersetzen. Ein zentrales Problem ist jedoch die geringe Effektivität
dieser Ansätze, die durch sofortiges Ausschwemmen und Absterben der transplantierten
Zellen erklärt werden. Eine temporäre Stimulation der Zellproliferation könnte eine
Verbesserung erbringen.
In dieser Arbeit wurde die Möglichkeit untersucht, die Proliferation von Kardiomyozyten mit
einem Tetrazyklin-stimuliertem System reversibel zu induzieren. Dazu wurde eine bereits
etablierte caERBB2-Zelllinie (Tre3G-caERBB2-GFP) verwendet, in der das Tet-On-System im
AAV1-Safe-Harbor-Lokus der ERC001-Zelllinie integriert wurde. Sie ermöglicht die Induktion
einer konstitutiv aktiven Variante des menschlichen ERBB2 durch Zugabe von Doxycyclin. Als
Kontrollzelllinie diente Tet-GFP (Tre3G-Tet-GFP) die nur GFP aber nicht caERBB2 exprimiert.
Beide Zelllinien reagierten auf Doxycyclin, was durch eine erhöhte GFP-Expression
nachweisbar war.
Mehrere Versuchsreihen verdeutlichten den Unterschied zwischen beiden Zelllinien. Bereits
nach wenigen Tagen der Doxycyclin-Behandlung konnte bei der caERBB2-Zelllinie ein
kompletter Verlust der physiologischen Kontraktionsfähigkeit beobachtet werden.
Histologische Analysen wiesen zu diesem Zeitpunkt einen signifikanten Anstieg EdU-positiver
Kerne im Vergleich zur Kontrollgruppe auf, was auf eine gesteigerte Zellproliferation
hindeutete. Diese Interpretation wurde durch einen deutlichen Anstieg des Gesamt-DNAGehalts
bestätigt. Die Sarkomerstruktur war, entsprechend der fehlenden
Kontraktionsfähigkeit, nicht mehr erkennbar. Trotzdem zeigten die behandelten EHTs einen
gegenüber unbehandelten Kontrollen gesteigerten Glukoseverbrauch, was auf eine deutliche
Stimulation von energieverbrauchenden Signalwegen hindeutet.
Interessanterweise zeigte sich drei bis vier Wochen nach gründlichem Auswaschen von
Doxycyclin nicht nur eine signifikante Zunahme der Kontraktionskraft in der Tre3G-caERBB2-
GFP-Zelllinie, sondern diese übertraf in allen Fällen die der unbehandelten caERBB2 Linie.
Zudem war ein gesteigerter Glukoseverbrauch, begleitet von einer histologisch nachweisbaren
Wiederherstellung der Sarkomerstruktur, feststellbar. Diese Beobachtungen deuten daauf hin,
dass die transiente Induktion von caERBB2 letztlich zur Ausbildung von mehr
kontraktionsfähigem Myokard führt.
Die Kontrollzelllinie zeigte erwartungsgemäß keine relevanten Veränderungen in Physiologie,
Anatomie oder Histologie. Lediglich die GFP-Expression war während und in geringerem
Ausmaß auch nach der Doxycyclin-Behandlung sichtbar. Eine leichte, temporäre Verringerung
der Kontraktionskraft wurde nach der Behandlung in dieser Zelllinie beobachtet, was
möglicherweise auf die toxischen Eigenschaften von Doxycyclin zurückzuführen ist.
Die Ergebnisse dieser Arbeit belegen und charakterisieren die Doxycyclin-regulierbare proproliferativen
Wirkung von caERBB2 in der modifizierten Tre3G-caERBB2-GFP-Zelllinie und
bilden die Grundlage für weiterführende Studien. Eine Übertragung der Resultate auf das
Tiermodell wird derzeit am Universitätsklinikum Hamburg-Eppendorf im Rahmen einer
präklinischen Studie untersucht
Test Beam Studies of Prototype Module and Test Measurements of MaPSA for The CMS Outer Tracker Phase-2 Upgrade
Get PDFThis thesis presents contributions to key aspects of the CMS Outer Tracker upgrade, including both 2S and PS module development. The first part focuses on the assembly and characterization of prototype 2S modules, with a sensor thickness of 240μm. The 2S prototype module demonstrated excellent performance, as validated through electrical and beam tests conducted at DESY, achieving a particle detection e ciency above 99.9% and a stub detection e ciency exceeding 99.7%. Beam tests also confirmed the design concept for discriminating low transverse momentum tracks, showing agreement with expectations.
The second part focuses on the development of a test stand for the quality assurance of the Macro Pixel Sub-Assembly (MaPSA), a main component of the PS module. Established to support the production of over 1250 PS modules at DESY, the test stand facilitates the testing and grading of MaPSAs. A comprehensive electrical testing workflow was implemented, including current measurements, pixel functionality tests, bump bonding evaluations, and threshold equalization. The threshold equalization procedure was opti- mized to achieve precise alignment of pixel responses, minimizing the risk of misclassifying untrimmable pixels. Through successful evaluations of prototype MaPSAs, the test stand has demonstrated its reliability and readiness for the production phase
Issue Tracking Ecosystems: Context and Best Practices
Get PDFIssue Tracking Systems (ITSs), such as GitHub and Jira, are popular tools that support Software Engineering (SE) organisations through the management of “issues”, which represent different SE artefacts such as requirements, development tasks, and maintenance items. ITSs also support internal linking between issues, and external linking to other tools and information sources. This provides SE organisations key forms of documentation, including forwards and backwards traceability (e.g., Feature Requests linked to sprint releases and code commits linked to Bug Reports). An Issue Tracking Ecosystem (ITE) is the aggregate of the central ITS and the related SE artefacts, stakeholders, and processes—with an emphasis on how these contextual factors interact with the ITS. The quality of ITEs is central to the success of these organisations and their software products. There are challenges, however, within ITEs, including complex networks of interlinked artefacts and diverse workflows. While ITSs have been the subject of study in SE research for decades, ITEs as a whole need further exploration.
In this thesis, I undertake the challenge of understanding ITEs at a broader level, addressing these questions regarding complexity and diversity. I interviewed practitioners and performed archival analysis on a diverse set of ITSs. These analyses revealed the context-dependent nature of ITE problems, highlighting the need for context-specific ITE research. While previous work has produced many solutions to specific ITS problems, these solutions are not consistently framed in a context-rich and comparable way, leading to a desire for more aligned solutions across research and practice. To address this emergent information and lack of alignment, I created the Best Practice Ontology for ITEs. Using this ontology, I curated a catalogue of Best Practices from existing literature, including Timely Severe Issue Resolution, Bug-to-Commit Linking, and Avoid Zombie Bugs. I also collected and created algorithms to automatically detect violations to these Best Practices. Finally, I proposed and evaluated tooling solutions that describe how to integrate these Best Practices into existing development environments.
The findings from this thesis enable a structured approach to improving the quality of ITEs. The Best Practice ontology, catalogue, and algorithms are contributions to researchers interested in understanding and improving ITEs. In practice, the context-aware catalogue and algorithms can be used to identify key areas for improvement, and to automate organisational processes such as maintaining a meaningful backlog and ensuring the completeness of issues.Issue Tracking Systems (ITSs) wie GitHub und Jira sind beliebte Tools, die Software Engineering (SE) Organisationen durch die Verwaltung von „Issues“ unterstützen, die verschiedene SE-Artefakte wie Anforderungen, Entwicklungsaufgaben und Wartungselemente darstellen. ITS unterstützen auch die interne Verknüpfung von Issues und die externe Verknüpfung mit anderen Tools und Informationsquellen. Dies bietet SE-Organisationen wichtige Formen der Dokumentation, einschließlich vorwärts und rückwärts gerichteter Rückverfolgbarkeit (z. B. Verknüpfung von Feature Requests mit Sprint Releases und Code Commits mit Bug Reports). Ein Issue Tracking Ecosystem (ITE) ist die Gesamtheit des zentralen ITS und der damit verbundenen SE-Artefakte, Stakeholder und Prozesse—mit dem Schwerpunkt darauf, wie diese kontextuellen Faktoren mit dem ITS interagieren. Die Qualität der ITEs ist entscheidend für den Erfolg dieser Organisationen und ihrer Softwareprodukte. Es gibt jedoch Herausforderungen innerhalb von ITEs, einschließlich komplexer Netzwerke miteinander verbundener Artefakte und unterschiedlicher Arbeitsabläufe. Während ITS seit Jahrzehnten Gegenstand der SE-Forschung sind, müssen ITEs als Ganzes weiter erforscht werden.
In dieser Arbeit stelle ich mich der Herausforderung, ITEs auf einer breiteren Ebene zu verstehen und diese Fragen hinsichtlich Komplexität und Vielfalt zu beantworten. Ich befragte Praktiker und führte eine Archivanalyse einer Vielzahl von ITS durch. Diese Analysen haben die kontextabhängige Natur von ITS-Problemen und -Lösungen aufgezeigt und den Bedarf an kontextspezifischer ITS-Forschung verdeutlicht. Während frühere Arbeiten vielfältige Analysen und Lösungen für spezifische ITS-Probleme hervorgebracht haben, sind diese Lösungen nicht durchgängig kontextabhängig und vergleichbar, was zu dem Wunsch nach besser abgestimmten Lösungen in Forschung und Praxis führt. Um dieses Informationsdefizit und den Mangel an Abstimmung zu beheben, habe ich eine ITE-Best-Practice-Ontologie erstellt. Anhand dieser Ontologie habe ich einen Katalog bewährter Verfahren aus der vorhandenen Literatur zusammengestellt. Anschließend habe ich Tooling-Lösungen vorgeschlagen und bewertet, die beschreiben, wie diese bewährten Verfahren in bestehende Umgebungen integriert werden können.
Die Erkenntnisse dieser Arbeit ermöglichen einen strukturierten Ansatz zur Verbesserung der ITE-Qualität. Die Ontologie, der Katalog und die Algorithmen sind Beiträge für Forscher, die sich für das ITE-Verständnis und die -Verbesserung interessieren. In der Praxis können der kontextbezogene Katalog und die Algorithmen genutzt werden, um Schlüsselbereiche für Verbesserungen zu identifizieren und organisatorische Prozesse zu automatisieren
Dynamics of driven antiferromagnetic skyrmions
Get PDFMagnetic Skyrmions are vortex-like quasi-particles stabilized by their topology and exist in several magnetic materials. While originally described by continuous vector fields that ensure topological protection, stable Skyrmions can also form in magnetic lattices, allowing for their experimental observation. Not least due to their potential applications in magnetic storage devices, Skyrmions have gained popularity since their first experimental observation in 2009. Ferromagnetic (FM) Skyrmions are nanometer-sized and can be driven by electric currents. They move as massless particles at an angle to the current flow and, therefore, exhibit the Skyrmion Hall effect. While ferromagnetic Skyrmions are still extensively studied in current research, the field of study evolve further to advanced structures. An emerging topic is the study of antiferromagnetic (AFM) Skyrmions. Unlike their ferromagnetic counterparts, AFM Skyrmions do not exhibit the Skyrmion Hall effect and can reach higher velocities when driven by electric currents, making them more attractive for technological applications. Moreover, AFM Skyrmions have a finite mass and exhibit behavior akin to classical particles, making them a fascinating subject for fundamental research.
Theoretical studies of AFM Skyrmions typically focuses on either synthetic antiferromagnets or G-type antiferromagnets, making it crucial to understand the distinctions between them. Synthetic antiferromagnets are a composition of two or more ferromagnetic layers that are coupled antiferromagnetically through a non-magnetic spacer layer. In contrast, G-type antiferromagnets exhibit a chessboard-like structure, where the sublattices are within the same layer. While Skyrmions have been experimentally observed in synthetic antiferromagnets, they remain elusive in other types of antiferromagnetic systems up to now. Both types are considered in this thesis. After introducing classical micromagnetic dynamics and general properties of a magnetic Skyrmion, we will discuss these two realizations of antiferromagnets. Furthermore, we will demonstrate that, independent of the type, the antiferromagnetic system can be treated as two effectively coupled ferromagnetic sublattices. In this thesis, the underlying mechanics of antiferromagnetic Skyrmion dynamics play a crucial role. To investigate them, we utilize the separation into two sublattices and treat the Skyrmions forming on these sublattices as rigid objects. This approach reveals that the driving mechanism of an AFM Skyrmion is due to a small displacement of its sublattice constituents. We develop a formalism that incorporates this mechanism and demonstrate that an AFM Skyrmion eventually mirrors the dynamics of a classical particle with finite mass. Furthermore, this framework allows us to fully characterize the resulting motion of an AFM Skyrmion driven by an external force. We apply this formalism to the case of current-driven Skyrmions, showing that it can predict the dynamics of an antiferromagnetic Skyrmion, regardless of the type of antiferromagnet, in various scenarios, even in fine detail. All results are compared to micromagnetic simulations.
As another possibility for driving AFM Skyrmions, we examine spin wave-driven Skyrmions in the second part of this thesis, focusing on a two-dimensional square lattice. We begin with formulating a classical spin wave theory explicitly tailored to this system by linearizing the equations of motion around a homogeneous ground state. This allows us to derive the dispersion relation and characterize different types of spin wave polarization, namely circularly and linearly polarized spin waves. All results are confirmed by simulations. Subsequently, we investigate the impact of spin waves on the Skyrmion. To do so, we simulate an isolated Skyrmion in the lattice and inject the spin wave via edge spin manipulation. Our observations reveal that spin waves generally accelerate the Skyrmion. While linearly polarized spin waves move the Skyrmion in the direction of wave propagation, circularly polarized spin waves induce an additional motion perpendicular to this direction, resulting in a Skyrmion Hall effect. The resulting Skyrmion acceleration depends on the properties of the spin wave, such as its wave number and amplitude. Furthermore, we investigate the impact of damping on spin waves and the resulting Skyrmion motion. We derive an expression for the decay of spin waves as they propagate through the lattice, focusing on the decay length. Additionally, we propose a concept for an antiferromagnetic Skyrmion racetrack that incorporates both spin wave decay and the Skyrmion Hall effect
Semiconductor Nanowire Field-Effect Transistors: Exploring Cation Exchange Mechanisms and Electrical Transport Properties
Get PDFModel-Based Deep Speech Enhancement for Improved Interpretability and Robustness
Get PDFTechnology advancements profoundly impact numerous aspects of life, including how we communicate and interact. For instance, hearing aids enable hearing-impaired or elderly people to participate comfortably in daily conversations; telecommunications equipment lifts distance constraints, enabling people to communicate remotely; smart machines are developed to interact with humans by understanding and responding to their instructions. These applications involve speech-based interaction not only between humans but also between humans and machines. However, the microphones mounted on these technical devices can capture both target speech and interfering sounds, posing challenges to the reliability of speech communication in noisy environments. For example, distorted speech signals may reduce communication fluency among participants during teleconferencing. Additionally, noise interference can negatively affect the speech recognition and understanding modules of a voice-controlled machine. This calls for speech enhancement algorithms to extract clean speech and suppress undesired interfering signals, improving the overall quality and intelligibility of speech.
Traditional speech enhancement algorithms often rely on simplifying assumptions, such as slowly changing noise, to estimate the parameters required for clean speech estimators. This may lead to less than satisfactory results in acoustically challenging scenarios. In recent years, the field has seen great strides through deep learning-based algorithms. The success of deep learning stems largely from its universal function approximation capability and scalability to large datasets. In particular, deep predictive approaches have received widespread attention due to their remarkable flexibility in incorporating key features of the target speech into various stages of the speech enhancement framework. These stages include input feature processing, network architecture design, training objective formulation, and optimization strategy development. Essentially, deep predictive methods aim to learn a mapping between noisy mixtures and clean speech by training deep neural networks (DNNs) on a large number of paired noisy-clean speech samples. However, the performance of these algorithms depends heavily on the quantity and diversity of training data. As a result, performance degradation often occurs when there is a data mismatch between training and testing, known as the generalization problem. Moreover, predictive approaches are typically framed as problems with a single output, which may result in erroneous estimates for complex and unseen samples without any indication of uncertainty. Indeed, due to the black-box nature of DNNs, deep learning-based algorithms produce clean speech estimates in a non-transparent manner, making them difficult to interpret. In this thesis, we aim to incorporate statistical models into DNN-based speech enhancement to improve its robustness and interpretability.
The first part of the thesis explores these ideas from the perspective of uncertainty. We augment predictive speech enhancement with an uncertainty estimation task, such that the network model can provide not only clean speech estimates but also their associated predictive uncertainty. Furthermore, since generic Bayesian methods for uncertainty modeling in deep learning usually involve costly sampling processes, this thesis seeks to leverage statistical knowledge from the speech processing domain to efficiently estimate uncertainty with minimal computational overhead. We experimentally demonstrate that the proposed uncertainty-augmented framework effectively identifies when predictions deviate significantly from the true data by producing large uncertainty estimates. This allows us to assess the model's confidence in predictions when clean speech ground truth is unavailable. Additionally, we show that the uncertainty-augmented methods grounded in statistical modeling improve speech enhancement performance compared to methods that predict a single filter mask only. Next, we explore the direct use of uncertainty estimates for speech enhancement tasks. This includes unsupervised domain adaptation, where we utilize uncertainty-based filtering to select high-quality pseudo-targets to alleviate generalization issues. In another application, alongside audio inputs, we further explore modeling uncertainty originating from distorted video signals in an audio-visual phoneme classification task and demonstrate how to exploit modality-wise uncertainty to achieve more effective and robust multimodal fusion.
In the second part of the thesis, we investigate the issues of interpretability and robustness by focusing on deep generative approaches. In contrast to predictive approaches that learn a deterministic mapping between noisy and clean speech, deep generative approaches aim to learn prior distributions of given data and reuse this knowledge to perform speech enhancement during inference. In the thesis, we consider a specific group of methods, which use a variational autoencoder (VAE) to learn a prior distribution of clean speech and combine it with an untrained non-negative matrix factorization (NMF)-based noise model to estimate a filter mask for speech enhancement. The statistically interpretable VAE-NMF framework exhibits an improved generalization ability to unseen acoustic conditions compared to predictive methods. However, training the VAE solely with clean speech makes it susceptible to noise interference during testing, especially for inputs with low signal-to-noise ratios. In this part, we aim to improve overall robustness in difficult acoustic conditions by augmenting separately the speech and noise models with noise information. The resulting noise-aware speech and noise models retain the high interpretability provided by statistical modeling while at the same time exhibiting improved speech enhancement performance in acoustically challenging environments