Statistical parametric speech synthesis (SPSS) has seen improvements over recent years, especially in terms of intelligibility. Synthetic speech is often clear and understandable, but it can also be bland and monotonous. Proper generation of natural speech prosody is still a largely unsolved problem. This is relevant especially in the context of expressive audiobook speech synthesis, where speech is expected to be fluid and captivating. In general, prosody can be seen as a layer that is superimposed on the segmental (phone) sequence. Listeners can perceive the same melody or rhythm in different utterances, and the same segmental sequence can be uttered with a different prosodic layer to convey a different message. For this reason, prosody is commonly accepted to be inherently suprasegmental. It is governed by longer units within the utterance (e.g. syllables, words, phrases) and beyond the utterance (e.g. discourse). However, common techniques for the modeling of speech prosody - and speech in general - operate mainly on very short intervals, either at the state or frame level, in both hidden Markov model (HMM) and deep neural network (DNN) based speech synthesis. This thesis presents contributions supporting the claim that stronger representations of suprasegmental variation are essential for the natural generation of fundamental frequency for statistical parametric speech synthesis. We conceptualize the problem by dividing it into three sub-problems: (1) representations of acoustic signals, (2) representations of linguistic contexts, and (3) the mapping of one representation to another. The contributions of this thesis provide novel methods and insights relating to these three sub-problems. In terms of sub-problem 1, we propose a multi-level representation of f0 using the continuous wavelet transform and the discrete cosine transform, as well as a wavelet-based decomposition strategy that is linguistically and perceptually motivated. In terms of sub-problem 2, we investigate additional linguistic features such as text-derived word embeddings and syllable bag-of-phones and we propose a novel method for learning word vector representations based on acoustic counts. Finally, considering sub-problem 3, insights are given regarding hierarchical models such as parallel and cascaded deep neural networks.

This thesis explores, through a number of production andperception studies, the nature of the voice source signal andhow it varies in spoken communication. Research is alsopresented that deals with the techniques and methodologies foranalysing and synthesising the voice source. The main analytictechnique involves interactive inverse filtering for obtainingthe source signal, which is then parameterised to permit thequantification of source characteristics. The parameterisationis carried by means of model matching, using the four-parameterLF model of differentiated glottal flow. The first three analytic studies focus on segmental andsuprasegmental determinants of source variation. As part of theprosodic variation of utterances, focal stress shows for theglottal excitation an enhancement between the stressed voweland the surrounding consonants. At a segmental level, the voicesource characteristics of a vowel show potentially majordifferences as a function of the voiced/voiceless nature of anadjacent stop. Cross-language differences in the extent anddirectionality of the observed effects suggest differentunderlying control strategies in terms of the timing of thelaryngeal and supralaryngeal gestures, as well as in thelaryngeal tensions settings. Different classes of voicedconsonants also show differences in source characteristics:here the differences are likely to be passive consequences ofthe aerodynamic conditions that are inherent to the consonants.Two further analytic studies present voice source correlatesfor six different voice qualities as defined by Laver'sclassification system. Data from stressed and unstressedcontexts clearly show that the transformation from one voicequality to another does not simply involve global changes ofthe source parameters. As well as providing insights into theseaspects of speech production, the analytic studies providequantitative measures useful in technology applications,particularly in speech synthesis. The perceptual experiments use the LF source implementationin the KLSYN88 synthesiser to test some of the analytic resultsand to harness them to explore the paralinguistic dimension ofspeech communication. A study of the perceptual salience ofdifferent parameters associated with breathy voice indicatesthat the source spectral slope is critically important andthat, surprisingly, aspiration noise contributes relativelylittle. Further perceptual tests using stimuli with differentvoice qualities explore the mapping between voice quality andits paralinguistic function of expressing emotion, mood andattitude. The results of these studies highlight the crucialrole of voice quality in expressing affect as well as providingpointers to how it combines withf0for this purpose. The last section of the thesis focuses on the techniquesused for the analysis and synthesis of the source. Asemi-automatic method for inverse filtering is presented, whichis novel in that it optimises the inverse filter by exploitingthe knowledge that is typically used by the experimenter whencarrying out manual interactive inverse filtering. A furtherstudy looks at the properties of the modified LF model in theKLSYN88 synthesiser: it highlights how it differs from thestandard LF model and discusses the implications forsynthesising the glottal source signal from LF model data.Effective and robust source parameterisation for the analysisof voice quality is the topic of the final paper: theeffectiveness of global, amplitude-based, source parameters isexamined across speech tokens with large differences inf0. Additional amplitude-based parameters areproposed to enable a more detailed characterisation of theglottal pulse. <b>Keywords:</b>Voice source dynamics, glottal sourceparameters, source-filter interaction, voice quality,phonation, perception, affect, emotion, mood, attitude,paralinguistic, inverse filtering, knowledge-based, formantsynthesis, LF model, fundamental frequency,f0.

This thesis consists of five articles and a summary. Thethesis deals with methods for measuring properties of thevoice. The methods are all computer-based, but utilisedifferent approaches for measuring different aspects of thevoice. Paper I introduces the Visual Sort and Rate (VSR) method forperceptual rating of voice quality. The method is based on theVisual Analogue Scale (VAS), but simultaneously shows allstimuli as icons along the VAS on the computer screen. As thelistener places similar-sounding stimuli close to each otherduring the rating process, comparing stimuli becomeseasier. Paper II introduces the correlogram. Fundamental frequencyF0 sometimes cannot be strictly defined, particularly forperturbed voice signals. The method displays multipleconsecutive correlation functions in a grey scale image. Thus,the correlogram avoids selecting a single F0 value. Rather itpresents an unbiased image of periodicity, allowing theinvestigator to select among several candidates, ifappropriate. PaperIII introduces a method for detection of phonation tobe utilised in voice accumulators. The method uses twomicrophones attached near the subject’s ears. Phase andamplitude relations of the microphone signals are used to forma phonation detector. The output of the method can be used tomeasure phonation time, speaking time and fundamental frequencyof the subject, as well as sound pressure level of both thesubject’s voicing and the ambient sounds. Paper IV introduces a method for Fourier analysis ofhigh-speed laryngoscopic imaging. The data from the consecutiveimages are re-arranged to form time-series that reflect thetime-variation of light intensity in each pixel. Each of thesetime series is then analysed by means of Fouriertransformation, such that a spectrum for each pixel isobtained. Several ways of displaying these spectra aredemonstrated. Paper V examines a test set-up for simultaneous recording ofairflow, intra-oral pressure, electro-glottography, audio andhigh-speed imaging. Data are analysed with particular focus onsynchronisation between glottal area and inverse filteredairflow. Several methodological aspects are also examined, suchas the difficulties in synchronising high-speed imaging datawith the other signals. / QC 20100609

There is a lot of research made on second language (L2) learning (Jesney, 2004). However the relationship between foreign accent, intelligibility and acoustics within pronunciation tutoring is quite an unresearched area (Thorén, 2008). The aim of the study was to analyze how a course in Swedish pronunciation affected foreign accent, intelligibility and acoustics among L2-learners of Swedish. A total of 41 people participated in the study: 16 L2-learners, 4 native Swedish speakers consisting a control group and 21 perceptual assessors with Swedish as native language. The L2-learners foreign accent and intelligibility were rated by the listeners before and after a course in Swedish pronunciation on an eight-point Likert-scale. The listeners also answered a questionnaire on factors potentially affecting the ratings. The acoustic measurements were made on ten words that were read aloud consisting of long and short allophones of five Swedish vowels. Formants, vowel duration and fundamental frequency were measured for the three closest and the three furthest from native pronunciation rated L2-learners. The results indicate that a class in Swedish pronunciation significantly decreased the L2-learners foreign accent. A strong correlation between the foreign accent and intelligibility ratings was found. Despite a strong correlation, no significant improvement concerning intelligibility could be established. The only factor that affected the intelligibility ratings were the assessor’s geographic affiliation. People from the western part of Sweden rated the intelligibility as less intelligible than raters from the eastern part of Sweden. The results from some of the acoustic measurements corresponded with the assessors ratings of foreign accent and intelligibility. The L2-learner rated as closest to native pronunciation was also the one with acousticly measured results close to the reference values regarding vowel duration (Elert, 1964; Gårding et al, 1974; Kügler, 2007; Thorén, i.d.) and fundamental frequency (Pegoraro Krook, 1988). The conclusion is that Swedish pronunciation tutoring should be focused on exercises that increase intelligibility because exercises that improve the foreign accent not necessarily increase the intelligibility. / Det finns en hel del forskning kring andraspråksinlärning (Jesney, 2004). Dock är förhållandet mellan brytning, förståelighet och akustiska variabler ett område som inte är lika väl beforskat inom uttalsundervisning i svenska (Thorén, 2008). Syftet med föreliggande studie var att undersöka hur en kurs i svenskt uttal påverkade brytning, förståelighet och akustiskt mätbara parametrar hos personer med svenska som andraspråk. Totalt medverkade 41 personer i studien varav 16 stycken var andraspråkstalare, 4 stycken var kontrollpersoner och resterande 21 var modersmålstalare av svenska som agerade lyssnarbedömare. Lyssnarbedömarnas skattningar baserades på två inspelningar av andraspråkstalarnas spontantal före och efter en kurs i svenskt uttal. Bedömningen gjordes på en åttagradig Likert-skala gällande grad av brytning samt förståelighet. Bedömarna fick även fylla i ett frågeformulär gällande vilka faktorer som potentiellt kunde påverka skattningen av brytning och förståelighet. De akustiska mätningarna gjordes på tio upplästa ord innehållande fem av svenskans vokaler. Formanter, vokalduration samt grundtonsfrekvens undersöktes hos sex andraspråkstalare, där de tre närmst respektive tre längst ifrån ett modersmålslikt svenskt uttal analyserades. Resultaten i föreliggande studie visade att kursen i svenskt uttal signifikant minskade andraspråkstalarnas brytning. Det fanns dessutom en tydlig korrelation mellan brytning- och förståelighetsskattningarna. Trots detta kunde ingen signifikant förbättring gällande kursdeltagarnas förståelighet påvisas. Den enda faktor som visade sig påverka förståelighetsskattningarna var bedömarnas geografiska tillhörighet där personer från västra Sverige bedömde förståeligheten som sämre jämfört med personer från östra Sverige. Resultaten från några av de akustiska mätningarna överensstämde med bedömarnas uppfattning av kursdeltagarnas grad av brytning och förståelighet. Det betydde att den som skattats närmast ett modersmålslikt svenskt uttal av lyssnarbedömarna också låg närmast referensvärdena vid den akustiska analysen gällande vokalduration (Elert, 1964; Gårding et al, 1974; Kügler, 2007; Thorén, i.d.) och grundtonsfrekvens (Pegoraro Krook, 1988). Slutsatsen i denna studie är att uttalsundervisning i svenska bör fokuseras på övningar som förbättrar förståelighet då övningar som förbättrar brytning inte nödvändigtvis gynnar förståeligheten.

<p>This thesis explores, through a number of production andperception studies, the nature of the voice source signal andhow it varies in spoken communication. Research is alsopresented that deals with the techniques and methodologies foranalysing and synthesising the voice source. The main analytictechnique involves interactive inverse filtering for obtainingthe source signal, which is then parameterised to permit thequantification of source characteristics. The parameterisationis carried by means of model matching, using the four-parameterLF model of differentiated glottal flow.</p><p>The first three analytic studies focus on segmental andsuprasegmental determinants of source variation. As part of theprosodic variation of utterances, focal stress shows for theglottal excitation an enhancement between the stressed voweland the surrounding consonants. At a segmental level, the voicesource characteristics of a vowel show potentially majordifferences as a function of the voiced/voiceless nature of anadjacent stop. Cross-language differences in the extent anddirectionality of the observed effects suggest differentunderlying control strategies in terms of the timing of thelaryngeal and supralaryngeal gestures, as well as in thelaryngeal tensions settings. Different classes of voicedconsonants also show differences in source characteristics:here the differences are likely to be passive consequences ofthe aerodynamic conditions that are inherent to the consonants.Two further analytic studies present voice source correlatesfor six different voice qualities as defined by Laver'sclassification system. Data from stressed and unstressedcontexts clearly show that the transformation from one voicequality to another does not simply involve global changes ofthe source parameters. As well as providing insights into theseaspects of speech production, the analytic studies providequantitative measures useful in technology applications,particularly in speech synthesis.</p><p>The perceptual experiments use the LF source implementationin the KLSYN88 synthesiser to test some of the analytic resultsand to harness them to explore the paralinguistic dimension ofspeech communication. A study of the perceptual salience ofdifferent parameters associated with breathy voice indicatesthat the source spectral slope is critically important andthat, surprisingly, aspiration noise contributes relativelylittle. Further perceptual tests using stimuli with differentvoice qualities explore the mapping between voice quality andits paralinguistic function of expressing emotion, mood andattitude. The results of these studies highlight the crucialrole of voice quality in expressing affect as well as providingpointers to how it combines with<i>f</i>₀for this purpose.</p><p>The last section of the thesis focuses on the techniquesused for the analysis and synthesis of the source. Asemi-automatic method for inverse filtering is presented, whichis novel in that it optimises the inverse filter by exploitingthe knowledge that is typically used by the experimenter whencarrying out manual interactive inverse filtering. A furtherstudy looks at the properties of the modified LF model in theKLSYN88 synthesiser: it highlights how it differs from thestandard LF model and discusses the implications forsynthesising the glottal source signal from LF model data.Effective and robust source parameterisation for the analysisof voice quality is the topic of the final paper: theeffectiveness of global, amplitude-based, source parameters isexamined across speech tokens with large differences in<i>f</i>₀. Additional amplitude-based parameters areproposed to enable a more detailed characterisation of theglottal pulse.</p><p><b>Keywords:</b>Voice source dynamics, glottal sourceparameters, source-filter interaction, voice quality,phonation, perception, affect, emotion, mood, attitude,paralinguistic, inverse filtering, knowledge-based, formantsynthesis, LF model, fundamental frequency,<i>f</i>₀.</p>

The term prosody refers to characteristics of speech such as intonation, timing, loudness, and other acoustical properties imposed by physical, intentional and emotional state of the speaker. Phone durations and fundamental frequency contours are considered as two of the most prominent aspects of prosody. Modeling phone durations and fundamental frequency contours in Turkish speech are studied in this thesis. Various methods exist for building prosody models. State-of-the-art is dominated by corpus-based methods. This study introduces corpus-based approaches using classification and regression trees to discover the relationships between prosodic attributes and phone durations or fundamental frequency contours. In this context, a speech corpus, designed to have specific phonetic and prosodic content has been recorded and annotated. A set of prosodic attributes are compiled. The elements of the set are determined based on linguistic studies and literature surveys. The relevances of prosodic attributes are investigated by statistical measures such as mutual information and information gain. Fundamental frequency contour and phone duration modeling are handled as independent problems. Phone durations are predicted by using regression trees where the set of prosodic attributes is formed by forward selection. Quantization of phone durations is studied to improve prediction quality. A two-stage duration prediction process is proposed for handling specific ranges of phone duration values. Scaling and shifting of predicted durations are proposed to minimize mean squared error. Fundamental frequency contour modeling is studied under two different frameworks. One of them generates a codebook of syllable-fundamental-frequency-contours by vector quantization. The codewords are used to predict sentence fundamental frequency contours. Pitch accent prediction by two different clustering of codewords into accented and not-accented subsets is also considered in this framework. Based on the experience, the other approach is initiated. An algorithm has been developed to identify syllables having perceptual prominence or pitch accents. The slope of fundamental frequency contours are then predicted for the syllables identified as accented. Pitch contours of sentences are predicted using the duration information and estimated slope values. Performance of the phone duration and fundamental frequency contour models are evaluated quantitatively using statistical measures such as mean absolute error, root mean squared error, correlation and by kappa coefficients, and by correct classification rate in case of discrete symbol prediction.

Orientador: Prof. Dr. Luiz Alberto Luz de Almeida / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Engenharia Elétrica, 2017. / O motor de indução é a máquina elétrica de maior utilização em todo o planeta e seu desempenho é fundamental nos processos produtivos, fazendo se necessário o funcionamento livre de falhas. Baseado na análise da assinatura da corrente do motor (MCSA) é possível apontar falhas em motores de indução, como barras quebradas, através da análise de variações na corrente do estator, que no domínio da frequência geram bandas laterais à frequência fundamental. Porém, devido à dificuldade e alta complexidade para se lidar com a grande diferença entre as magnitudes das bandas laterais e a frequência fundamental, foi proposto na literatura uma técnica que atenua a componente da frequência fundamental via Transformada Recursiva Discreta de Fourier (RDFT) com objetivo de amplificar os espectros de bandas laterais gerados. Entretanto, a técnica proposta estima a componente fundamental baseando-se em uma frequência fixa (60Hz), sem considerar as oscilações presentes na rede que podem diretamente afetar o resultado da atenuação. É proposto neste trabalho uma filtragem adaptativa híbrida analógico-digital para melhoria na atenuação da componente fundamental através da implementação de um sistema compensador das oscilações da rede composto por um estimador de frequência do tipo "Zero-Crossing" e um oscilador controlado numericamente (NCO). Isto acarreta em baixa complexidade, aumentando a eficiência e confiabilidade do controle dos dados e acima de tudo levando em conta o contexto atual de redução de custos, permite a portabilidade para sistemas de baixo custo e Iot. / The induction motor is the most applied electrical machine around the planet and in its majority, plays a fundamental role in the productive process, requiring faults free functioning. Based on motor current signature analysis (MCSA) it is possible point faults in induction motors, as broken bars, through the analysis of the stators current imbalances, which in frequency domain generate sidebands around the fundamental frequency. Nevertheless, due the difficulty and the high complexity to handle the differences between the sidebands and fundamental frequency magnitudes, a technique which suppresses the fundamental frequency via Recursive Discrete Fourier Transform was proposed in order to amplify the sidebands spectrum generated. However, the proposed technique estimates the fundamental component based on a fixed frequency (60Hz), without considering the grid oscillations which can directly affect the result of the fundamental attenuation. It is proposed in this study a hybrid analogic-digital adaptive filtering in order to improve the fundamental component cancelling technique by implementing a grid oscillations compensator system composed by a Zero-Crossing Frequency Estimator and a Numerically Controlled Oscillator (NCO). It will result in low complexity, increasing the data control efficiency and reliability and above all taking in consideration the current reduction cost context, allow the portability to low cost and Iot systems.

Ce travail de thèse interroge le lien qui existe entre voix et genre. Le triple dispositif analytique sociophonétique, consistant à articuler données phonétiques, expérimentales et ethnographiques, a permis d’étudier comment une voix est perçue comme genrée et comment des locutrices/eurs utilisent des pratiques vocales pour indexer des identités de genre. Deux expériences dans lesquelles étaient utilisés comme stimuli des voix de synthèse et des voix resynthétisées ont permis d’observer que la fréquence fondamentale et les fréquences de résonance jouent des rôles différents dans la perception du genre. Une troisième expérience avec des voix de locutrices/eurs trans (transgenres, transsexuel-le-s) a permis de reproduire les résultats des deux expériences précédentes : en deçà d’un certain seuil de fréquence fondamentale, les voix tendent à être perçues comme « voix d’hommes » ; la perception genrée de voix produites avec des fréquences fondamentales supérieures à ce seuil est cependant largement déterminée par les fréquences de résonance.L’étude de pratiques vocales utilisées par des locutrices/eurs trans a soulevé un ensemble de questions sur le passing de genre et sur la co-indexation d’identités et de postures par la voix. Elle a aussi soulevé la question de la légitimité de chercheurs identifiés comme hommes cisgenres à réaliser ce type d’étude. Une démarche ethnographique a pu apporter des éléments de réponse à ces différentes questions. Une analyse de la littérature phonétique a finalement permis de montrer que celle-ci, à travers ses questions et hypothèses de recherche, ses axiomes, ses analyses et interprétations des données, peut véhiculer une idéologie de genre binaire et sexiste. / The aim of this dissertation is to investigate the relationship between voice and gender. Phonetic, experimental and ethnographic data have been used to study how the voice is perceived as gendered and how speakers use vocal practices to index gender identities. Two experiments with synthetized and resynthesized voices have shown that fundamental frequency and resonance frequencies play different roles in the perception of gender. The results of these experiments could be reproduced in a third experiment with voices of transgender speakers: under a certain fundamental frequency threshold, voices tend to be perceived as “male voices”; but above this threshold, resonance frequencies define if the voice is perceived as “female voice” or “male voice”. The study of the vocal practices of transgender speakers raised questions about gender passing, and about the indexical link between identities, stances and voice. It also raised the question of the legitimacy of researchers that are identified as cisgender males to do research on trans speaker voices. These different questions could be addressed through ethnographic data. Finally, an analysis of the phonetic literature showed that the research questions and hypotheses, the axioms, the analyses and interpretations of data one can find in phonetic studies can be a vehicle for a sexist and binary gender ideology.

This master thesis concerns with emotional states and gender recognition on the basis of speech signal analysis. We used various prosodic and cepstral features for the description of the speech signal. In the text we describe non-invasive methods for glottal pulses estimation. The described features of speech were implemented in MATLAB. For their classification we used the GMM classifier, which uses the Gaussian probability distribution for modeling a feature space. Furthermore, we constructed a system for recognition of emotional states of the speaker and a system for gender recognition from speech. We tested the success of created systems with several features on speech signal segments of various lengths and compared the results. In the last part we tested the influence of speaker and gender on the success of emotional states recognition.

The modal and spectral characteristic belongs among important human acoustic spaces of vocal tract. They occur at generating vowels and other acoustic aspects of human speech. We can observe the resonant phenomena of acoustic cavity of vocal tract in the human speech spectrum, primary however at vowels generation. However near vocal tract occurs series of frequency tops in the spectrum of vowels, which necessarily may not be resonant origin. That is why sometimes quite difficult assign is right frequency tops to resonant tops of acoustic cavity. It consist in operate of acoustic excitation of vocal tracts. The pronounced of vowels loudly and in a whisper has different excitation of vocal tract. At generating vowels loudly is excited by scheme of harmonic components outspread to fundamental frequency of glottis. At talking in a whisper is vocal tract excited by continuous spectrum generated by turbulent fluxion of exhaled flatus over glottis. We give a name "formant" to a frequency, at which happens to resonance of acoustic space. Aim of this work is analysis of Czech vowels formants generated loudly and in a whisper. Experimental metering of these formants was performed on human vocalic tract for all vowels. Further then on artificially created vocalic tracts for vowels A, I. Then were modal characteristics of vocal cavity for vowels A, I, tested by method of final elements with the help of computing program ANSYS. In this work were surveyed courses of acoustic pressures for individual formants, influence sizes vocal tract and influence of correct mouth opening on formants. Also has been effected computational simulation of harmonic excitation on tract by side of glottis.