Ces vingt dernières années, la grande majorité des travaux réalisés dans le domaine de l’analyse de la qualité a été consacrée à la quantification de la distortion engendrée par le traitement d’une image. Par conséquent, l’image originale était toujours considérée comme étant de la meilleure qualité possible. Dans ce genre de scénario, la notion de qualité peut être exprimée comme la fidélité de la version traitée à sa version de référence. Cependant, des algorithmes de post-traitement permettent d’ajuster les propriétés esthétiques d’une image afin d’améliorer la qualité perceptible. Dans ce cas, il n’existe pas d’image ayant la meilleure qualité possible, et l’approche classique basée sur la fidélité ne peut plus être utilisée. L’objectif de cette thèse est de corriger les méthodologies d’analyse de la qualité afin de résoudre les difficultés d’évaluation de qualité que soulève le post-traitement. Les algorithmes de post-traitement, en rapport avec le sujet de cette thèse, proviennent de deux groupes : l’amélioration d’image, caractérisée par l’accentuation du contraste, et les techniques de compression de la plage dynamique (également appelée mappage tonal). Les méthodologies de l’analyse de qualité applicables dans ces domaines, tant subjectives qu’objectives, y sont étudiées, et les solutions proposées permettent de surpasser les méthodes les plus récentes. De plus, une nouvelle méthodologie est présentée afin d’évaluer les performances des indicateurs de la qualité objective, corrigeant les défauts des méthodes actuellement utilisées. / The vast majority of the work done in the field of quality assessment during last two decades has been dedicated to the quantification of the distortion caused by the processing of an image. The original image was, therefore, always considered to be of the best possible quality. In this kind of scenario, the notion of quality can be expressed as the fidelity of the processed version to the reference. However, some post-processing algorithms enable to adjust aesthetic properties of an image in order to enhance the perceived quality. In such cases, the best possible quality image is not available and the classical fidelity approach is no longer applicable. The goal of this thesis is to revise the quality assessment methodologies to cope with the challenges brought by the post-processing into the quality evaluation. The post-processing algorithms, relevant to the topic of this thesis, come from two groups – image enhancement, represented by image sharpening, and dynamic range compression (also known as tone-mapping) techniques. Both subjective and objective quality assessment methodologies applicable in these areas are studied and the suitable solutions, outperforming the state-of-the-art methods, are proposed. Moreover, a novel methodology for evaluating the performance of objective quality metrics, overcoming the shortcomings of the currently available methods, is presented.

Lexical tone plays an important role in tonal languages. Acoustically, pitch is determined by the periodicity of speech, which is measured as the fundamental frequency (F0) of acoustic signals. In each tonal language, there are a certain number of lexical tones that are described by distinctive pitch contours. Cantonese and Mandarin have four and six tones, respectively. / People with sensorineural hearing loss have difficulty in utilizing spectral information for speech recognition and rely heavily on temporal information. The temporal information of speech is divided into three parts, based on the rate of amplitude fluctuation: temporal envelope (below 50 Hz), periodicity (50-500 Hz), and fine structure (above 500 Hz). / Previous studies show that hearing-impaired people have difficulties in perceiving tones, even though they are aided with cochlear implants (CIs). In this thesis, two approaches are investigated to improve Chinese tone recognition. In the first approach, TEPCs go through a process of non-linear expansion in order to increase the modulation depth of periodicity-related amplitude fluctuation. Results of listening tests show that TEPC expansion leads to a noticeable improvement on tone identification accuracy. In the second approach, the effectiveness of enhancing temporal periodicity cues in noise is investigated. Temporal periodicity cues are simplified into a sinusoidal wave with frequency equivalent to the F0 of speech. This leads to a consistent and significant improvement on tone identification performance at different noise levels. This part of research is expected to be helpful in designing CI processing strategy for effective speech perception of tonal languages. / The goals of this thesis are to investigate what are the effective temporal cues for lexical tone perception of Chinese and how to manipulate or enhance these cues for better performance of tone perception. We adopt the research method of acoustic simulation with normal-hearing subjects. A four-channel noise-excited vocoder is used to generate test stimuli for tone identification. / We compare the contributions of temporal envelope and periodicity components (TEPCs) from different frequency regions to tone recognition in Cantonese and Mandarin. It is observed that TEPCs from high-frequency region (1-4 kHz) are more important than those from low-frequency region (< 1 kHz). In noise condition, tone recognition performance with temporal cues degrades and more spectral information is needed. / Yuan, Meng. / Adviser: Tan Lee. / Source: Dissertation Abstracts International, Volume: 71-01, Section: B, page: 0565. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2009. / Includes bibliographical references (leaves 131-146). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. Ann Arbor, MI : ProQuest Information and Learning Company, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts in English and Chinese.

人类如何在颜色、视觉对象和听觉对象存在很大差异的情况下实现感知恒定，这是认知神经科学的一个根本问题。大脑处理差异的一个重要办法是依靠背景环境，为感知颜色、视觉对象和听觉对象提供参照。在语音感知中，在言语信号中存在话者间以及话者内差异的情况下实现语音恒定也是很关键的。根据语境归一化机制，听者可以通过语境（即目标语音周围的其他语音）适应话者的语音空间。在本论文中，我以声调为例考察了语音恒定的问题。 / 第一，在一个跨语言研究中，我发现不同语音系统的结构会影响听者听辨多个话者所发的声调。普通话听者可以不靠语境准确地听辨多个话者所发的声调，而广东话听者则被多个话者间的音高差异误导。这一不同可以归根于广东话声调系统中存在多个调型相同的平调。这一发现有助理解不同语音系统的结构对于话者差异的影响的抵抗力。 / 第二，我发现而广东话听者需要语境以估计某一话者的单调范围来帮助判断平调。带有相同基频信息的言语语境和非言语语境的作用不同。非言语语境的作用很小，而言语语境，无论是否有语义，都有作用，不过有语义的言语语境作用更大。 / 关于语境归一化的神经基础，我最早在N400 时间窗（250-500 毫秒）发现了归一化效应。这说明言语语境提供话者的音高参照以准确地分析词的语音特征，帮助词义提取。当归一化以一种自上而下的方式实施时，归一化的发生不晚于音素加工阶段（PMN，220-350 毫秒）。这些探索性的EEG 研究是最早考察语境归一化的神经基础的研究。 / 第三，我提出了一个关于语音在大脑中表征的混合模型，以整合两种对立的观点。这个模型中，低层表征是实际听到的不同话者所说的语音的例子，高层表征则是反映不同话者间语音相似性的更为抽象的表征。我找到了一些初步的支持证据，比如语音辨认的准确性与话者的音高在群体分布中的典型性有显著相关。这说明高层的表征形式是由一个语言社区中话者音高的总体分布所决定的。这一模型需要被进一步检验，比如通过语言习得研究，考察习得新语音范畴时低层和高层表征的动态发展。 / 总之，本论文对于理解语音系统的结构有帮助，并且阐明了语境归一化的机制和神经基础以及语音的多层表征形式。但是还有很多未解决的问题有待进一步考察。 / How humans achieve constancy in the perception of color, visual object and auditory object despite the tremendous variation is a fundamental question in cognitive neuroscience. An important way that the brain tackles variation is through reliance on the context, which provides a reference for the perception of an object. In speech perception, it is critical to achieve phonetic constancy above the inter- and intra-talker variation in speech signals. According to the context-dependent normalization mechanism, listeners adapt to a talker’s phonetic space via the context (i.e., neighboring sounds of a speech sound to be recognized). This contextually built phonetic space serves as a reference for compensating for talker variation. In this thesis, I have examined the question of phonetic constancy using lexical tones as a case study. / Firstly, in a cross-linguistic study, I found that the structure of phonological inventories influences categorization of multi-talker tone stimuli. Mandarin listeners correctly categorized multi-talker stimuli without contexts, whereas Cantonese listeners were misguided by acoustic variation between talkers, a difference attributable to the existence of multiple level tones with a similar F0 contour in Cantonese. This finding has implications for understanding the structure of phonological inventories in terms of resistance to talker variability. / Secondly, I found that Cantonese listeners could resolve the ambiguity of level tones by adapting to talker-specific pitch references via a context. Speech and nonspeech contexts contribute unequally to talker adaptation. Nonspeech contexts have a minimal effect, whereas speech contexts, no matter meaningful or not, facilitate adaptation, but congruent semantic content further enlarges the facilitatory effect. / As for the neural locus of context-dependent normalization, I found normalization effects in the N400 time-window (250-500 ms). It indicates that speech contexts facilitate retrieval of semantic memory, by providing talker-specific references to accurately assess the phonetic property of a word. When implemented in a top-down manner, context-dependent normalization occurs no later than the phonemic level of processing (Phonological Mapping Negativity, 220-350 ms). These EEG studies, though exploratory, are among the first to examine the neural processes of context-dependent normalization. / Thirdly, I proposed a hybrid model of mental representations to reconcile two opposite views. In this model, at a lower level are encountered exemplars of speech sounds from different talkers, and at a higher level are abstract representations that reflect the general similarity of speech sounds across talkers. I found initial evidence for this model, such as a significant correlation between the identification accuracy and the typicality of a talker’s pitch range in the population distribution, which suggests that higher-level representations are shaped by the global distribution of talkers’ vocal characteristics in a community. This model needs to be further tested in studies on language learning to examine dynamic development of talker-specific and abstract representations for new phonological categories. / In conclusion, this thesis has implications for understanding the structure of phonological inventories in the world’s languages; it also sheds light on mechanisms and neural processes of context-dependent normalization and the hybrid nature of mental representations. Many unresolved questions remain to be examined in future studies. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Zhang, Caicai. / Thesis (Ph.D.) Chinese University of Hong Kong, 2014. / Includes bibliographical references (leaves 173-194). / Abstracts also in Chinese. / Zhang, Caicai.

No description available.

This thesis studies the interactions of vowel tone with consonantal voice. Briefly, tone-voice interactions refer to: (i) voiced--not voiceless--onsets block high tone spreading; (ii) voiceless--not voiced--onsets block low tone spreading; (iii) sonorant onsets are transparent to both tonal processes. There are many variations to these archetypical patterns of tone-voice interactions. I argue that these variations as well as the archetypical patterns can receive a revealing account from the phonological theory. Specifically, this thesis explores the Prosodic Hypothesis of Tone-Voice, which claims: (i) tone must be represented prosodically (namely, tone is associated to mora); and (ii) tone-voice relations must be expressed by conditions (namely, path conditions, proposed in Archangeli and Pulleyblank (in prep)). By exploiting tonal representations and conditions on tone-voice, the Prosodic Hypothesis provides a principled account of tone-voice in Ngizim, Ewe, and Nupe. The result is a principled theory that unifies the known phonetic and phonological facts about tone-voice and that makes testable predictions about the nature and type of expected tone-voice interactions. In addition to tone-voice, this thesis investigates a range of theoretical issues from tonal representations, to onset representations, to the privative voicing theory to Grounded Phonology (Archangeli and Pulleyblank in prep.). I demonstrate that detailed formal analyses of tone-voice can not only uncover facts about tone-voice, but can also make important contributions to phonological theory.

No description available.

Wong, Ying Wai. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2007. / Includes bibliographical references (leaves 56-60). / Abstracts in English and Chinese. / Acknowledgement --- p.i / Abstract --- p.iii / 摘要 --- p.iv / Contents --- p.v / Chapter Chapter 1 --- Introduction --- p.1 / Chapter Chapter 2 --- Tones in Continuous Speech --- p.4 / Chapter 2.1 --- Production of tones --- p.4 / Chapter 2.2 --- Perception of tones --- p.7 / Chapter Chapter 3 --- Tone Production --- p.9 / Chapter 3.1 --- Tones in isolation - Experiment 1 --- p.9 / Introduction --- p.10 / Material --- p.10 / Recording and F0 extraction --- p.10 / Results and discussion --- p.11 / Chapter 3.2 --- Tones in context - Experiment 2 --- p.13 / Introduction --- p.14 / Material --- p.15 / Recording and F0 extraction --- p.16 / Results - Anticipatory effect --- p.17 / Results - Carryover effect --- p.18 / Results - Pitch target computation --- p.20 / Results - Magnitude of variation --- p.21 / Chapter 3.3 --- Discussion --- p.22 / Relative magnitude of contextual variations --- p.23 / Cantonese T4 in context --- p.24 / Further generalizations --- p.25 / Chapter 3.4 --- Summary of findings --- p.27 / Chapter Chapter 4 --- Tone Perception --- p.28 / Chapter 4.1 --- Temporal distribution of tonal information - Expt. 3 and 4 --- p.29 / Experiment 3 - Stimulus recording --- p.29 / Experiment 3 - Stimulus preparation --- p.31 / Experiment 3 - Procedure --- p.31 / Experiment 3 - Results --- p.32 / Experiment 4 - Stimulus preparation --- p.33 / Experiment 4 - Procedure --- p.34 / Experiment 4 - Results --- p.34 / Chapter 4.2 --- Effect of intensity 一 Experiment 5 --- p.36 / Stimulus preparation --- p.37 / Procedure --- p.38 / Results - Tone identification responses --- p.38 / Results - Effect of intensity --- p.40 / Chapter 4.3 --- Contour tone perception - Experiment 6 --- p.43 / Stimulus preparation --- p.43 / Procedure --- p.44 / Results --- p.44 / Chapter 4.4 --- Summary of findings --- p.46 / Chapter Chapter 5 --- Conclusions and Discussion --- p.47 / Chapter 5.1 --- Tone Production --- p.47 / Typology of tone rules --- p.48 / Acquisition of tone production --- p.49 / Chapter 5.2 --- Tone Perception --- p.50 / Evidence from Speech Production Phenomena --- p.50 / Role of the first half --- p.52 / Acquisition of tone perception --- p.53 / Bibliography --- p.56

Thesis (Ph.D.)--University of Delaware, 1997. / Professor in charge: William J. Idsardi. Includes bibliographical references.

published_or_final_version / Linguistics / Master / Master of Arts

No description available.