Towards speaking style transplantation in speech synthesis

One of the biggest challenges in speech synthesis is the production of naturally sounding synthetic voices. This means that the resulting voice must be not only of high enough quality but also that it must be able to capture the natural expressiveness imbued in human speech. This paper focus on solving the expressiveness problem by proposing a set of different techniques that could be used for extrapolating the expressiveness of proven high quality speaking style models into neutral speakers in HMM-based synthesis. As an additional advantage, the proposed techniques are based on adaptation approaches, which means that they can be used with little training data (around 15 minutes of training data are used in each style for this paper). For the final implementation, a set of 4 speaking styles were considered: news broadcasts, live sports commentary, interviews and parliamentary speech. Finally, the implementation of the 5 techniques were tested through a perceptual evaluation that proves that the deviations between neutral and speaking style average models can be learned and used to imbue expressiveness into target neutral speakers as intended.

Analysis of pixel-mapping rounding on geometric distortion as a prediction for view synthesis distortion

We analyze the performance of the geometric distortion, incurred when coding depth maps in 3D Video, as an estimator of the distortion of synthesized views. Our analysis is motivated by the need of reducing the computational complexity required for the computation of synthesis distortion in 3D video encoders. We propose several geometric distortion models that capture (i) the geometric distortion caused by the depth coding error, and (ii) the pixel-mapping precision in view synthesis. Our analysis starts with the evaluation of the correlation of geometric distortion values obtained with these models and the actual distortion on synthesized views. Then, the different geometric distortion models are employed in the rate-distortion optimization cycle of depth map coding, in order to assess the results obtained by the correlation analysis. Results show that one of the geometric distortion models is performing consistently better than the other models in all tests. Therefore, it can be used as a reasonable estimator of the synthesis distortion in low complexity depth encoders.

Synthesis and biological activity of DNA damaging agents that form decoy binding sites for the estrogen receptor

It is a goal of cancer chemotherapy to achieve the selective killing of tumor cells while minimizing toxicity to normal tissues. We describe the design of selective toxins forming DNA adducts that attract the estrogen receptor (ER), a transcription factor that is overexpressed in many human breast and ovarian tumors. The compounds consist of 4-(3-aminopropyl)-N,N-(2-chloroethyl)-aniline linked to 2-(4′-hydroxyphenyl)-3-methyl-5-hydroxy-indole. The former moiety is a DNA damaging nitrogen mustard and the latter is a ligand for the ER. The connection between these groups was refined to permit DNA adducts formed by the mustard portion of the molecule to present the ligand domain so that it was able to interact efficiently with the ER. By using 16-mers containing specific DNA adducts, it was determined that monoadducts and putative intrastrand crosslinks were preferred targets for the ER over interstrand crosslinks. A series of structurally related 2-phenylindole mustards was prepared, some of which were selectively toxic to the ER-positive breast cancer cell line MCF-7, as compared with the ER(−) negative line MDA-MB231. The ability both to bind to DNA and to interact significantly with the ER were essential to achieve selective lethality toward ER(+) cells. Compounds forming DNA adducts without the ability to bind receptor showed similar toxicities in the two cell lines. Several models could explain the selective toxicity of the mustard–phenylindole compounds toward ER(+) cells. The favored model suggests that a mustard–DNA adduct is shielded by the ER from DNA repair enzymes and hence cells possessing an abundance of the ER selectively retain the adduct and are killed.

Computer speech synthesis: its status and prospects.

Computer speech synthesis has reached a high level of performance, with increasingly sophisticated models of linguistic structure, low error rates in text analysis, and high intelligibility in synthesis from phonemic input. Mass market applications are beginning to appear. However, the results are still not good enough for the ubiquitous application that such technology will eventually have. A number of alternative directions of current research aim at the ultimate goal of fully natural synthetic speech. One especially promising trend is the systematic optimization of large synthesis systems with respect to formal criteria of evaluation. Speech recognition has progressed rapidly in the past decade through such approaches, and it seems likely that their application in synthesis will produce similar improvements.

Modified PNA design and synthesis: a novel approach using Molecular Dynamics and Metadynamics

Gli acidi peptido nucleici sono potenti strumenti utilizzati in ambito biotecnologico per colpire DNA o RNA. PNA contenenti basi o backbone modificati sono attualmente studiati per migliorarne le proprietà in ambito biologico. Bersagliare i micro RNA (anti-miR) è particolarmente interessante nell’ottica di future applicazioni terapeutiche, ma strumenti computazionali che aiutino nel design di nuovi PNA anti-miR non sono stati ancora completamente sviluppati. Le proprietà conformazionali del singolo filamento di PNA (non modificato o recante modificazioni in γ) e dei duplex PNA:RNA e i processi di re-annealing e melting sono stati studiati tramite Dinamica Molecolare e Metadinamica. L’approccio computazionale consolidato, assieme a un programma modificato per la generazione delle strutture dei duplex contenenti PNA, è stato utilizzato per il virtual screening di PNA contenenti basi modificate. Sono state inoltre sintetizzate le unità per l’ottenimento del composto più promettente e una funzione idrolitica da legare al monomero finale.

Integration of different models in the design of chemical processes: Application to the design of a power plant

With advances in the synthesis and design of chemical processes there is an increasing need for more complex mathematical models with which to screen the alternatives that constitute accurate and reliable process models. Despite the wide availability of sophisticated tools for simulation, optimization and synthesis of chemical processes, the user is frequently interested in using the ‘best available model’. However, in practice, these models are usually little more than a black box with a rigid input–output structure. In this paper we propose to tackle all these models using generalized disjunctive programming to capture the numerical characteristics of each model (in equation form, modular, noisy, etc.) and to deal with each of them according to their individual characteristics. The result is a hybrid modular–equation based approach that allows synthesizing complex processes using different models in a robust and reliable way. The capabilities of the proposed approach are discussed with a case study: the design of a utility system power plant that has been decomposed into its constitutive elements, each treated differently numerically. And finally, numerical results and conclusions are presented.

Population synthesis of isolated neutron stars with magneto-rotational evolution

We revisit the population synthesis of isolated radio-pulsars incorporating recent advances on the evolution of the magnetic field and the angle between the magnetic and rotational axes from new simulations of the magneto-thermal evolution and magnetosphere models, respectively. An interesting novelty in our approach is that we do not assume the existence of a death line. We discuss regions in parameter space that are more consistent with the observational data. In particular, we find that any broad distribution of birth spin periods with P0 ≲ 0.5 s can fit the data, and that if the alignment angle is allowed to vary consistently with the torque model, realistic magnetospheric models are favoured compared to models with classical magneto-dipolar radiation losses. Assuming that the initial magnetic field is given by a lognormal distribution, our optimal model has mean strength 〈log B0[G]〉 ≈ 13.0–13.2 with width σ(log B0) = 0.6–0.7. However, there are strong correlations between parameters. This degeneracy in the parameter space can be broken by an independent estimate of the pulsar birth rate or by future studies correlating this information with the population in other observational bands (X-rays and γ-rays).

Population synthesis of isolated neutron stars with magneto-rotational evolution – II. From radio-pulsars to magnetars

Population synthesis studies constitute a powerful method to reconstruct the birth distribution of periods and magnetic fields of the pulsar population. When this method is applied to populations in different wavelengths, it can break the degeneracy in the inferred properties of initial distributions that arises from single-band studies. In this context, we extend previous works to include X-ray thermal emitting pulsars within the same evolutionary model as radio-pulsars. We find that the cumulative distribution of the number of X-ray pulsars can be well reproduced by several models that, simultaneously, reproduce the characteristics of the radio-pulsar distribution. However, even considering the most favourable magneto-thermal evolution models with fast field decay, lognormal distributions of the initial magnetic field overpredict the number of visible sources with periods longer than 12 s. We then show that the problem can be solved with different distributions of magnetic field, such as a truncated lognormal distribution, or a binormal distribution with two distinct populations. We use the observational lack of isolated neutron stars (NSs) with spin periods P > 12 s to establish an upper limit to the fraction of magnetars born with B > 1015 G (less than 1 per cent). As future detections keep increasing the magnetar and high-B pulsar statistics, our approach can be used to establish a severe constraint on the maximum magnetic field at birth of NSs.

Mandibular Torus : A Synthesis of New and Previously Reported Data and a Discusssion of its Cause

Usinig original data on 1,5000 mandibles, but mainly previously published data, I present a overview of the distribution characteristics of mandibular torus and a hypothesis concerning its cause. Pedigree studies have established that genetic factors influence torus development. Extrinsic factors are strongly implicated by other evidence: prevalence among Arctic peoples, effect of dietary change, age regression, preponderance in males and on the right side, effect of cranial deformation, concurrence with palatine torus and maxillary alveolar exostoses, and clinical evidence. I propose that the primary factor is masticatory stress. According to a mechanism suggested by orthodontic research, the horizontal component of bite force tips the lower canine, premolars and first molar so that their root apices exert pressure on the periodontal membrane, causing formation of new bone on the lingual cortical plate of the alveolar process. Thus formed, the hyperostosis is vulnerable to trauma and its periosteal covering becomes bruised causing additional deposition of bone. Genes influence torus indirectly through their effect on occlusion. A patern of increased expressivity with incidence suggests that a quasicontinuous model may provide a better fit to pedigree data than single locus models previously tested.

Learning Models of Gene Expression from Synthetic DNA Sequences

Thesis (Ph.D.)--University of Washington, 2016-06

A convergent solution-phase synthesis of the macrocycle Ac-Phe-[Orn-Pro-D-Cha-Trp-Arg], a potent new antiinflammatory drug

Relatively few cyclic peptides have reached the pharmaceutical marketplace during the past decade, most produced through fermentation rather than made synthetically. Generally, this class of compounds is synthesized for research purposes on milligram scales by solid-phase methods, but if the potential of macrocyclic peptidomimetics is to be realized, low-cost larger scale solution-phase syntheses need to be devised and optimized to provide sufficient quantities for preclinical, clinical, and commercial uses. Here, we describe a cheap, medium-scale, solution-phase synthesis of the first reported highly potent, selective, and orally active antagonist of the human C5a receptor. This compound, Ac-Phe[Orn-Pro-D-Cha-Trp-Arg], known as 3D53, is a macrocyclic peptidomimetic of the human plasma protein C5a and displays excellent antiinflammatory activity in numerous animal models of human disease. In a convergent approach, two tripeptide fragments Ac-Phe-Orn-(Boc)-Pro-OH and H-D-Cha-Trp(For)-Arg-OEt were first prepared by high-yielding solution-phase couplings using a mixed anhydride method before coupling them to give a linear hexapeptide which, after deprotection, was obtained in 38% overall yield from the commercially available amino acids. Cyclization in solution using BOP reagent gave the antagonist in 33% yield (13% overall) after HPLC purification. Significant features of the synthesis were that the Arg side chain was left unprotected throughout, the component Boe-D-Cha-OH was obtained very efficiently via hydrogenation Of D-Phe with PtO2 in TFA/water, the tripeptides were coupled at the Pro-Cha junction to minimize racemization via the oxazolone pathway, and the entire synthesis was carried out without purification of any intermediates. The target cyclic product was purified (>97%) by reversed-phase HPLC. This convergent synthesis with minimal use of protecting groups allowed batches of 50100 g to be prepared efficiently in high yield using standard laboratory equipment. This type of procedure should be useful for making even larger quantities of this and other macrocyclic peptidomimetic drugs.

Robustness of mathematical models for biological systems

The robustness of mathematical models for biological systems is studied by sensitivity analysis and stochastic simulations. Using a neural network model with three genes as the test problem, we study robustness properties of synthesis and degradation processes. For single parameter robustness, sensitivity analysis techniques are applied for studying parameter variations and stochastic simulations are used for investigating the impact of external noise. Results of sensitivity analysis are consistent with those obtained by stochastic simulations. Stochastic models with external noise can be used for studying the robustness not only to external noise but also to parameter variations. For external noise we also use stochastic models to study the robustness of the function of each gene and that of the system.

Different routes to pure food allergens: extraction, recombinant techniques and total chemical synthesis for obtaining the plant-food pan-allergen LTP

La richiesta di allergeni puri è in continuo aumento per scopi diagnostici, come standard per metodi di rilevamento e di quantificazione, per l'immunoterapia e per lo studio a livello molecolare dei meccanismi delle reazioni allergiche, al fine di facilitare lo sviluppo di possibili cure. In questa tesi di dottorato sono descritte diverse strategie per l’ottenimento di forme pure di non-specific Lipid Transfer Proteins (nsLTPs), le quali sono state riconosciute essere rilevanti allergeni alimentari in molti frutti e verdure comunemente consumati e sono state definite come modello di veri allergeni alimentari. Una LTP potenzialmente allergenica, non nota in precedenza, è stata isolata dalle mandorle, mentre una LTP dall’allergenicità nota contenuta nelle noci è stata prodotta mediante tecniche di DNA ricombinante. Oltre a questi approcci classici, metodi per la sintesi chimica totale di proteine sono stati applicati per la prima volta alla produzione di un allergene, utilizzando Pru p 3, la LTP prototipica e principale allergene della pesca nell'area mediterranea, come modello. La sintesi chimica totale di proteinepermette di controllarne completamente la sequenza e di studiare la loro funzione a livello atomico. La sua applicazione alla produzione di allergeni costituisce perciò un importante passo avanti nel campo della ricerca sulle allergie alimentari. La proteina Pru p 3 è stata prodotta nella sua intera lunghezza e sono necessari solo due passaggi finali di deprotezione per ottenere il target nella sua forma nativa. Le condizioni sperimentali per tali deprotezioni sono state messe a punto durante la produzione dei peptidi sPru p 3 (1-37) e sPru p 3 (38-91), componenti insieme l'intera proteina. Tecniche avanzate di spettrometria di massa sono state usate per caratterizzare tutti i composti ottenuti, mentre la loro allergenicità è stata studiata attraverso test immunologici o approcci in silico.

Design and synthesis of potential antimicrobial agents

Chorismate mutase is one of the essential enzymes in the shikimate pathway and is key to the survival of the organism Mycobacterium tuberculosis. The x-ray crystal structure of this enzyme from Mycobacterium tuberculosis was manipulated to prepare an initial set of in silico protein models of the active site. Known inhibitors of the enzyme were docked into the active site using the flexible ligand / flexible active site side chains approach implemented in CAChe Worksystem (Fujitsu Ltd). The resulting complexes were refined by molecular dynamics studies in explicit water using Amber 9. This yielded a further set of protein models that were used for additional rounds of ligand docking. A binding hypothesis was established for the enzyme and this was used to screen a database of commercially available drug-like compounds. From these results new potential ligands were designed that fitted appropriately into the active site and matched the functional groups and binding motifs founds therein. Some of these compounds and close analogues were then synthesized and submitted for biological evaluation. As a separate part of this thesis, analogues of very active anti-tuberculosis pyridylcarboxamidrazone were also prepared. This was carried out by the addition and the deletion of the substitutions from the lead compound thereby preparing heteroaryl carboxamidrazone derivatives and related compounds. All these compounds were initially evaluated for biological activity against various gram positive organisms and then sent to the TAACF (USA) for screening against Mycobacterium tuberculosis. Some of the new compounds proved to be at least as potent as the original lead compound but less toxic.

Expert systems for fault tree synthesis

Hazard and operability (HAZOP) studies on chemical process plants are very time consuming, and often tedious, tasks. The requirement for HAZOP studies is that a team of experts systematically analyse every conceivable process deviation, identifying possible causes and any hazards that may result. The systematic nature of the task, and the fact that some team members may be unoccupied for much of the time, can lead to tedium, which in turn may lead to serious errors or omissions. An aid to HAZOP are fault trees, which present the system failure logic graphically such that the study team can readily assimilate their findings. Fault trees are also useful to the identification of design weaknesses, and may additionally be used to estimate the likelihood of hazardous events occurring. The one drawback of fault trees is that they are difficult to generate by hand. This is because of the sheer size and complexity of modern process plants. The work in this thesis proposed a computer-based method to aid the development of fault trees for chemical process plants. The aim is to produce concise, structured fault trees that are easy for analysts to understand. Standard plant input-output equation models for major process units are modified such that they include ancillary units and pipework. This results in a reduction in the nodes required to represent a plant. Control loops and protective systems are modelled as operators which act on process variables. This modelling maintains the functionality of loops, making fault tree generation easier and improving the structure of the fault trees produced. A method, called event ordering, is proposed which allows the magnitude of deviations of controlled or measured variables to be defined in terms of the control loops and protective systems with which they are associated.