Thesis (MScEng)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: Several accents of English are spoken in South Africa. Automatic speech recognition (ASR) systems should therefore be able to process the di erent accents of South African English (SAE). In South Africa, however, system development is hampered by the limited availability of speech resources. In this thesis we consider di erent acoustic modelling approaches and system con gurations in order to determine which strategies take best advantage of a limited corpus of the ve accents of SAE for the purpose of ASR. Three acoustic modelling approaches are considered: (i) accent-speci c modelling, in which accents are modelled separately; (ii) accent-independent modelling, in which acoustic training data is pooled across accents; and (iii) multi-accent modelling, which allows selective data sharing between accents. For the latter approach, selective sharing is enabled by extending the decision-tree state clustering process normally used to construct tied-state hidden Markov models (HMMs) by allowing accent-based questions. In a rst set of experiments, we investigate phone and word recognition performance achieved by the three modelling approaches in a con guration where the accent of each test utterance is assumed to be known. Each utterance is therefore presented only to the matching model set. We show that, in terms of best recognition performance, the decision of whether to separate or to pool training data depends on the particular accents in question. Multi-accent acoustic modelling, however, allows this decision to be made automatically in a data-driven manner. When modelling the ve accents of SAE, multi-accent models yield a statistically signi cant improvement of 1.25% absolute in word recognition accuracy over accent-speci c and accentindependent models. In a second set of experiments, we consider the practical scenario where the accent of each test utterance is assumed to be unknown. Each utterance is presented simultaneously to a bank of recognisers, one for each accent, running in parallel. In this setup, accent identi cation is performed implicitly during the speech recognition process. A system employing multi-accent acoustic models in this parallel con guration is shown to achieve slightly improved performance relative to the con guration in which the accents are known. This demonstrates that accent identi cation errors made during the parallel recognition process do not a ect recognition performance. Furthermore, the parallel approach is also shown to outperform an accent-independent system obtained by pooling acoustic and language model training data. In a nal set of experiments, we consider the unsupervised reclassi cation of training set accent labels. Accent labels are assigned by human annotators based on a speaker's mother-tongue or ethnicity. These might not be optimal for modelling purposes. By classifying the accent of each utterance in the training set by using rst-pass acoustic models and then retraining the models, reclassi ed acoustic models are obtained. We show that the proposed relabelling procedure does not lead to any improvements and that training on the originally labelled data remains the best approach. / AFRIKAANSE OPSOMMING: Verskeie aksente van Engels word in Suid Afrika gepraat. Outomatiese spraakherkenningstelsels moet dus in staat wees om verskillende aksente van Suid Afrikaanse Engels (SAE) te kan hanteer. In Suid Afrika word die ontwikkeling van spraakherkenningstegnologie egter deur die beperkte beskikbaarheid van geannoteerde spraakdata belemmer. In hierdie tesis ondersoek ons verskillende akoestiese modelleringstegnieke en stelselkon gurasies ten einde te bepaal watter strategie e die beste gebruik maak van 'n databasis van die vyf aksente van SAE. Drie akoestiese modelleringstegnieke word ondersoek: (i) aksent-spesi eke modellering, waarin elke aksent apart gemodelleer word; (ii) aksent-onafhanklike modellering, waarin die akoestiese afrigdata van verskillende aksente saamgegooi word; en (iii) multi-aksent modellering, waarin data selektief tussen aksente gedeel word. Vir laasgenoemde word selektiewe deling moontlik gemaak deur die besluitnemingsboom-toestandbondeling-algoritme, wat gebruik word in die afrig van gebinde-toestand verskuilde Markov-modelle, uit te brei deur aksent-gebaseerde vrae toe te laat. In 'n eerste stel eksperimente word die foon- en woordherkenningsakkuraathede van die drie modelleringstegnieke vergelyk in 'n kon gurasie waarin daar aanvaar word dat die aksent van elke toetsspraakdeel bekend is. In hierdie kon gurasie word elke spraakdeel slegs gebied aan die modelstel wat ooreenstem met die aksent van die spraakdeel. In terme van herkenningsakkuraathede, wys ons dat die keuse tussen aksent-spesi eke en aksent-onafhanklike modellering afhanklik is van die spesi eke aksente wat ondersoek word. Multi-aksent akoestiese modellering stel ons egter in staat om hierdie besluit outomaties op 'n data-gedrewe wyse te neem. Vir die modellering van die vyf aksente van SAE lewer multi-aksent modelle 'n statisties beduidende verbetering van 1.25% absoluut in woordherkenningsakkuraatheid op in vergelyking met aksent-spesi eke en aksent-onafhanklike modelle. In 'n tweede stel eksperimente word die praktiese scenario ondersoek waar daar aanvaar word dat die aksent van elke toetsspraakdeel onbekend is. Elke spraakdeel word gelyktydig gebied aan 'n stel herkenners, een vir elke aksent, wat in parallel hardloop. In hierdie opstelling word aksentidenti kasie implisiet uitgevoer. Ons vind dat 'n stelsel wat multi-aksent akoestiese modelle in parallel inspan, e ense verbeterde werkverrigting toon in vergelyking met die opstelling waar die aksent bekend is. Dit dui daarop dat aksentidenti seringsfoute wat gemaak word gedurende herkenning, nie werkverrigting be nvloed nie. Verder wys ons dat die parallelle benadering ook beter werkverrigting toon as 'n aksent-onafhanklike stelsel wat verkry word deur akoestiese en taalmodelleringsafrigdata saam te gooi. In 'n nale stel eksperimente ondersoek ons die ongekontroleerde herklassi kasie van aksenttoekennings van die spraakdele in ons afrigstel. Aksente word gemerk deur menslike transkribeerders op grond van 'n spreker se moedertaal en ras. Hierdie toekennings is nie noodwendig optimaal vir modelleringsdoeleindes nie. Deur die aksent van elke spraakdeel in die afrigstel te klassi seer deur van aanvanklike akoestiese modelle gebruik te maak en dan weer modelle af te rig, word hergeklassi seerde akoestiese modelle verkry. Ons wys dat die voorgestelde herklassi seringsalgoritme nie tot enige verbeterings lei nie en dat dit die beste is om modelle op die oorspronklike data af te rig.

published_or_final_version / English Studies / Master / Master of Arts

published_or_final_version / abstract / Linguistics / Master / Master of Arts in Applied Linguistics

This thesis examines existing theories of Spanish stress assignment in generative phonology and proposes an alternative theory that is more effective in predicting the surface representations of Spanish stress. Stress is characterized according to traditional textbook standards and examples are given (Chapter I). The current theoretical setting, especially the theories of James W. Harris, is then described (Chapter II). This writer's own theory, based upon an underlying distinction between tense and lax vowels, is delineated (Chapter III) and defended (Chapter IV). The new stress assignment rule--along with a rule of vowel laxing before a word boundary (#) and a rule of stress adjustment--shows stress in Spanish to be predictable and, therefore, not phonemic.

Communication in a second or multiple languages has become essential in the globalized world. However, acquiring a second language (L2) after a critical period is universally acknowledged to be challenging (Lenneberg, 1967). Late learners hardly reach a nativelike level in L2, particularly in its pronunciation, and their incomplete phonological acquisition is manifested by a foreign accent—a common and persistent feature of otherwise fluent L2 speech. Although foreign-accented speech is widespread, it has been a target of social constraints in L2-speaking communities, causing many learners and instructors to seek out ways to reduce foreign accents. Accordingly, research in L2 speech has unceasingly examined various learner-external and learner-internal factors of the occurrence of foreign accents as well as nonnative speech characteristics underlying the judgment of the degree of foreign accents. The current study aimed to expand the understanding of the characteristics and judgments of foreign accents by investigating phonological awareness, a construct pertinent to learners’ phonological knowledge, which has received little attention in research on foreign accents. The current study was exploratory and non-experimental research that targeted 40 adults with Korean-accented English living in the United States. The study first examined how 23 raters speaking American English as their native language detect, perceive, describe, and rate Korean-accented English. Through qualitative and quantitative analyses of the accent perception data, the study identified various phonological and phonetic deviations from the nativelike sounds, which largely result from the influence of first language (Korean) on L2 (English). The study then probed the relationship between foreign accents and learners’ awareness of the phonological system of L2, which was measured using production, perception, and verbalization tasks that tapped into the knowledge of L2 phonology. The study found a significant inverse relationship between the degree of a foreign accent and phonological awareness, particularly implicit knowledge of L2 segmentals. Further in-depth analyses revealed that explicit knowledge of L2 phonology alone was not sufficient for targetlike pronunciation. Findings suggest that L2 speakers experience varying degrees of difficulty in perceiving and producing different L2 segmentals, possibly resulting in foreign-accented speech.

Trust and speech are both essential aspects of human interaction. On the one hand, trust is necessary for vocal communication to be meaningful. On the other hand, humans have developed a way to infer someone’s trustworthiness from their voice, as well as to signal their own. Yet, research on trustworthiness attributions to speakers is scarce and contradictory, and very often uses explicit data, which do not predict actual trusting behaviour. However, measuring behaviour is very important to have an actual representation of trust. This thesis contains 5 experiments aimed at examining the influence of various voice characteristics — including accent, prosody, emotional expression and naturalness — on trusting behaviours towards virtual players and robots. The experiments have the "investment game"—a method derived from game theory, which allows to measure implicit trustworthiness attributions over time — as their main methodology. Results show that standard accents, high pitch, slow articulation rate and smiling voice generally increase trusting behaviours towards a virtual agent, and a synthetic voice generally elicits higher trustworthiness judgments towards a robot. The findings also suggest that different voice characteristics influence trusting behaviours with different temporal dynamics. Furthermore, the actual behaviour of the various speaking agents was modified to be more or less trustworthy, and results show that people’s trusting behaviours develop over time accordingly. Also, people reinforce their trust towards speakers that they deem particularly trustworthy when these speakers are indeed trustworthy, but punish them when they are not. This suggests that people’s trusting behaviours might also be influenced by the congruency of their first impressions with the actual experience of the speaker’s trustworthiness — a "congruency effect". This has important implications in the context of Human–Machine Interaction, for example for assessing users’ reactions to speaking machines which might not always function properly. Taken together, the results suggest that voice influences trusting behaviour, and that first impressions of a speaker’s trustworthiness based on vocal cues might not be indicative of future trusting behaviours, and that trust should be measured dynamically.

by Lui Ling-yee. / Thesis (M.A.)--Chiense University of Hong Kong, 1981. / Bibliography: leaves 94-97.

Sin Ping Wong. / Thesis (M.Phil.)--Chinese University of Hong Kong, 1991. / Bibliography: leaves 143-147. / Abstract --- p.i / Acknowledgements --- p.ii / Table of Contents --- p.iii / Chapter CHAPTER 0. --- INTRODUCTION --- p.1 / Chapter 0.1. --- Learning the English Sound System --- p.1 / Chapter 0.2. --- The Importance of the Acquisition of English Word Stress --- p.5 / Chapter 0.2.1. --- "Stress, Morphology and Syntax" --- p.5 / Chapter 0.2.2. --- Stress and Intelligibility of Non-native Speech --- p.7 / Chapter CHAPTER 1. --- THE STRESS PATTERNS OF ENGLISH WORDS --- p.12 / Chapter 1.1. --- The Phonetic Properties of English Word Stress --- p.12 / Chapter 1.2. --- Theoretical Approaches to English Word Stress --- p.13 / Chapter 1.2.1. --- Chomsky and Halle (1968) 一一 The Sound Pattern of English --- p.14 / Chapter 1.2.1.1. --- Strong/Weak Syllables --- p.15 / Chapter 1.2.1.2. --- The Transformational Cycle --- p.15 / Chapter 1.2.1.3. --- The Main Stress Rule --- p.18 / Chapter 1.2.1.4. --- The Compound Stress Rule and Nuclear Stress Rule --- p.20 / Chapter 1.2.2. --- Liberman and Prince (1977) -- Metrical Phonology --- p.22 / Chapter 1.2.2.1. --- The Stress Rules --- p.24 / Chapter 1.2.2.2. --- The Lexical Category Prominence Rule --- p.26 / Chapter 1.2.3. --- Selkirk (1980) ´ؤ Prosodic Categories --- p.29 / Chapter 1.2.4. --- Hayes (1981) ´ؤ Extrametricality --- p.34 / Chapter 1.2.4.1. --- Rime Projection --- p.34 / Chapter 1.2.4.2. --- Extrametricality --- p.35 / Chapter 1.2.4.3. --- English Stress Rules and Word Tree Construction --- p.36 / Chapter 1.3. --- Prerequisites for the Acquisition of English Word Stress --- p.40 / Chapter CHAPTER 2. --- STUDIES ON THE ACQUISITION OF STRESS --- p.43 / Chapter 2.1. --- The Acquisition of Stress by Children --- p.43 / Chapter 2.2. --- Adults' Stress Patterns - Native Speakers and ESL Learners --- p.45 / Chapter CHAPTER 3. --- METHODOLOGY --- p.51 / Chapter 3.1. --- Research Objectives --- p.51 / Chapter 3.2. --- Research Design --- p.55 / Chapter 3.2.1. --- Test Materials --- p.55 / Chapter 3.2.1.1. --- Criteria for the Coinage of Test Items --- p.55 / Chapter 3.2.1.2. --- The Rime Structure of Test Words --- p.57 / Chapter 3.2.1.3. --- Number of Syllables and Directionality of Rule Application --- p.59 / Chapter 3.2.1.4. --- Syntactic Categories --- p.61 / Chapter 3.2.2. --- Subjects --- p.64 / Chapter 3.2.3. --- Procedure --- p.64 / Chapter CHAPTER 4. --- FINDINGS --- p.68 / Chapter 4.1. --- Overall Performance --- p.69 / Chapter 4.1.1. --- Proficiency Test --- p.69 / Chapter 4.1.2. --- Overall Performance in Test Items --- p.70 / Chapter 4.1.2.1. --- Secondary Students Versus University Students --- p.72 / Chapter 4.1.2.2. --- Syntactic Category --- p.73 / Chapter 4.2. --- Learners' Stress Patterns Classified According to Syllable Structures --- p.76 / Chapter 4.2.1. --- The Stress Patterns of ESL Learners in Verbs --- p.78 / Chapter 4.2.1.1. --- 2-syllable Verbs --- p.78 / Chapter 4.2.1.2. --- 3-syllable Verbs (Regularly Patterned Responses) --- p.83 / Chapter 4.2.1.3. --- 3-syllable Verbs (Irregularly Patterned Responses) --- p.87 / Chapter 4.2.2. --- Stress Patterns of ESL Learners in Nouns --- p.92 / Chapter 4.2.2.1. --- 2-syllable Nouns --- p.92 / Chapter 4.2.2.2. --- 3-syllable Nouns --- p.96 / Chapter 4.3. --- 4-Syllable Nouns --- p.104 / Chapter CHAPTER 5. --- DISCUSSION --- p.116 / Chapter 5.1. --- Summary of Findings --- p.117 / Chapter 5.2. --- The Metrical Theory and the Stress Patterns of ESL --- p.123 / Chapter 5.2.1. --- Rime Structures --- p.123 / Chapter 5.2.2. --- Extrametricality --- p.125 / Chapter 5.2.2.1. --- Consonant Extrametricality --- p.125 / Chapter 5.2.2.2. --- Noun Extrametricality --- p.126 / Chapter 5.2.3. --- Directionality --- p.130 / Chapter 5.2.4. --- Stress Assignment --- p.132 / Chapter 5.3. --- Second Language Acquisition --- p.137 / Chapter 5.3.1. --- Learners' Strategy --- p.137 / Chapter 5.3.2. --- Group Difference --- p.138 / Chapter 5.3.3. --- L1 Transfer --- p.139 / Chapter CHAPTER 6. --- CONCLUSION --- p.141 / REFERENCES --- p.143 / Appendix A --- p.i / Appendix B --- p.vi / Appendix C --- p.xvi

Thesis (Ph.D.)--Massachusetts Institute of Technology, Dept. of Linguistics and Philosophy, 1980. / MICROFICHE COPY AVAILABLE IN ARCHIVES AND HUMANITIES. / Bibliography: leaves 246-253. / by Janet Breckenridge Pierrehumbert. / Ph.D.

This study investigates the acquisition of English lexical stress by Cantonese-English bilingual children and compares the bilinguals with English monolingual children of the same age. The aim of this study is to find out how Cantonese-English bilingual children develop English lexical stress under the possible interaction of the two phonological systems. Two age groups were included: 2;06 and 3;0. Two acoustic correlates for lexical stress were measured: syllable duration and peak F0. Words in two language contexts were examined: disyllabic words in English speech and code-switched English words in Cantonese speech. / Data of eight bilingual children and seven monolingual children were used. All the bilingual data came from CHILDES database and monolingual data came from multiple sources. Disyllabic words were extracted from English utterances, and the stressed/unstressed syllable duration and peak F0 ratios were calculated to see how contrastive the stress patterns were. Results indicated that the monolingual children displayed a more contrastive stress pattern in both acoustic correlates, although syllable duration is found to be more reliable than peak F0 in stress distinction. Stress pattern in code-switched words were also analyzed. It was found that bilingual children displayed a stronger pitch difference in code-switched words, more contrastive than the patterns in their English speech. The results suggest that between 2;06 and 3;0, Cantonese-English bilingual children used less duration and pitch contrast for lexical stress than English monolingual children did, and they were slower in developing native-like lexical stress pattern during this period. On the other hand, Cantonese-English bilingual children used acoustic properties for lexical stress differently in the two language contexts, indicating that they possessed some kind of metalinguistic awareness at such an early age. / 本研究探索粵語和英語雙語兒童對英語詞重音的早期習得情況，並將雙語兒童與英語單語兒童進行對比討論。研究目的旨在探討雙語語音系統之間是否存在相互影響，以及此影響如何改變雙語兒童的語言韻律發展。研究對象包括處於2 歲6個月及3 歲這兩個年齡階段的粵英雙語及英語單語兒童。文章選取兒童自然對話中,位於句中位置的雙音節詞，對其音節時長和音高峰值進行測量。此外，本研究發現雙語兒童存在代碼轉換的現象，即在粵語表達中插入英語辭彙，因此本文同時也對被插入的英語辭彙的詞重音進行探究，以期對詞重音的雙語習得獲得更詳盡的瞭解。 / 本研究所用數據來自八名粵英雙語兒童和七名英語單語兒童。雙語兒童數據取自“兒童語言資料交換系統”（CHILDES）中的“香港雙語兒童語料庫”，單語兒童數據則來自多方資料。研究將所提取的雙音節詞的重音音節與非重音音節的時長及音高峰值分別測量之後，計算出重音音節與非重音音節的比值。比值越大表明兒童在產出中越能明顯區別重音音節與非重音音節，即對英語詞重音的掌握越好。 / 研究結果發現單語兒童的表現領先于雙語兒童，體現在單語兒童可以將重音與非重音音節的時長與音高明顯區分，而雙語兒童產出的雙音節詞中，重音音節和非重音音節在時長和音高方面的區別都不明顯。然而對於被插入粵語句子中的英語雙音節詞，雙語兒童卻能很好地區分開重音音節和非重音音節的音高 。 / 研究結果表明，在2 歲6 個月至3 歲之間的這段時間，粵語和英語雙語兒童對英語詞重音的習得逊於英語單語兒童。但另一方面，粵語和英語雙語兒童表現出對語言環境的敏感，可以根據語言環境和交流對象的不同改變對語音特徵的運用，展現出一定程度的元（後設）語言覺識。 / Li, Jingwen. / Thesis M.Phil. Chinese University of Hong Kong 2014. / Includes bibliographical references (leaves 56-65). / Abstracts also in Chinese. / Title from PDF title page (viewed on 11, October, 2016). / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only.