Aplicação de uma rede neural artificial para a avaliação da rugosidade e soprosidade vocal

A avaliação perceptivo-auditiva tem papel fundamental no estudo e na avaliação da voz, no entanto, por ser subjetiva está sujeita a imprecisões e variações. Por outro lado, a análise acústica permite a reprodutibilidade de resultados, porém precisa ser aprimorada, pois não analisa com precisão vozes com disfonias mais intensas e com ondas caóticas. Assim, elaborar medidas que proporcionem conhecimentos confiáveis em relação à função vocal resulta de uma necessidade antiga dentro desta linha de pesquisa e atuação clínica. Neste contexto, o uso da inteligência artificial, como as redes neurais artificiais, indica ser uma abordagem promissora. Objetivo: Validar um sistema automático utilizando redes neurais artificiais para a avaliação de vozes rugosas e soprosas. Materiais e métodos: Foram selecionadas 150 vozes, desde neutras até com presença em grau intenso de rugosidade e/ou soprosidade, do banco de dados da Clínica de Fonoaudiologia da Faculdade de Odontologia de Bauru (FOB/USP). Dessas vozes, 23 foram excluídas por não responderem aos critérios de inclusão na amostra, assim utilizaram-se 123 vozes. Procedimentos: avaliação perceptivo-auditiva pela escala visual analógica de 100 mm e pela escala numérica de quatro pontos; extração de características do sinal de voz por meio da Transformada Wavelet Packet e dos parâmetros acústicos: jitter, shimmer, amplitude da derivada e amplitude do pitch; e validação do classificador por meio da parametrização, treino, teste e avaliação das redes neurais artificiais. Resultados: Na avaliação perceptivo-auditiva encontrou-se, por meio do teste Coeficiente de Correlação Intraclasse (CCI), concordâncias inter e intrajuiz excelentes, com p = 0,85 na concordância interjuízes e p variando de 0,87 a 0,93 nas concordâncias intrajuiz. Em relação ao desempenho da rede neural artificial, na discriminação da soprosidade e da rugosidade e dos seus respectivos graus, encontrou-se o melhor desempenho para a soprosidade no subconjunto composto pelo jitter, amplitude do pitch e frequência fundamental, no qual obteve-se taxa de acerto de 74%, concordância excelente com a avaliação perceptivo-auditiva da escala visual analógica (0,80 no CCI) e erro médio de 9 mm. Para a rugosidade, o melhor subconjunto foi composto pela Transformada Wavelet Packet com 1 nível de decomposição, jitter, shimmer, amplitude do pitch e frequência fundamental, no qual obteve-se 73% de acerto, concordância excelente (0,84 no CCI), e erro médio de 10 mm. Conclusão: O uso da inteligência artificial baseado em redes neurais artificiais na identificação, e graduação da rugosidade e da soprosidade, apresentou confiabilidade excelente (CCI > 0,80), com resultados semelhantes a concordância interjuízes. Dessa forma, a rede neural artificial revela-se como uma metodologia promissora de avaliação vocal, tendo sua maior vantagem a objetividade na avaliação.

A nasalidade no dialeto quilombola do Norte de Minas: uma análise contrastiva baseada em corpus

This research investigated the nasality of vowels in the spontaneous speech of inhabitants of the quilombola communities of Brejo dos Crioulos and Poções (MG). As a theoretical framework, we based on the assumptions of Phonetics and Phonology, in renowned scholars on the investigation of nasality (CAGLIARI, 1977; CÂMARA JR., 1984, 2013; BISOL, 2013; ABAURRE; PAGOTTO, 1996; SILVA, 2015), with subsidies of the Corpus Linguistics. Its general goal was to investigate the occurrence of nasality, in the dialect of these quilombola communities, and their linguistic behavior, considering the linguistic factors that can interfere in the phenomenon. Specifically it was aimed to a) detect the occurrence of nasalized vowels with the help of the resources that the Corpus Linguistics provides (Praat and WorldSmith Tolls); b) discriminate the different types of occurring contexts of nasalized vowels; c) make quantitative and qualitative analyzes of the nasalized vowels in the study corpus; d) describe and analyze the behavior of nasalized vowels and; e) contrast the values of F1 and F2 of the oral and nasalized vowels. It was hypothesized that the nasality happens because it is conditioned by the nasal segment following the nasalized vowel - phonological process of “assimilation” - its position as the primary stress and grammatical category. It was believed that the quilombolas communities of Brejo dos Crioulos and Poções produce nasalized vowels in their speech and this linguistic phenomenon is favored by the adjacent presence of consonants or nasal vowels. Furthermore, it was hypothesized that the values of F1 and F2 of oral and nasalized vowels in these communities are distinct. The following research questions were elaborated: (i) is the presence of nasalized vowels in the speech of these quilombola communities conditioned to the presence of a nasal sound segment? (ii) does the nasal sound segment following the nasalized vowel favor the occurrence of the nasality phenomenon? is there a difference between the values of F1 and F2 of the oral and nasalized vowels in both quilombola communities considered? To compose our corpus, 24 interviews recordings were used (12 female speakers and 12 male speakers), a total of 24 participants. It was found that the following nasal sound segment tends to condition the nasalized vowel. In general, it assimilates the lowering of the soft palate of nasal consonant segment immediately following, but there are cases of nasal vowel segment - regressive assimilation; the stressed syllable tends to favor the nasality, but it occurs in pretonic and postonic position as well; F1 and F2 values of oral and nasalized vowels in the quilombola communities of Poções and Brejo dos Crioulos are distinct: the group of Brejo dos Crioulos tends to produce the F1 of oral and nasalized vowels more lowered than the group of Poções and the F2, in a more anterior position. The nasality tends to occur in verbs and nouns, although it is not specific to a grammatical category. This research found cases of spurious nasalization, confirming previous studies. In turn, it revealed cases of lexical items with favorable context for nasalization, but with its non-occurrence. This last case, considered as the lowering of the uniform soft palate in PB, presented pronounced vowels without the soft palate lowering. That is, it was detected variation in the phenomenon of nasalization in PB. With this work, it was promoted the discussion about nasality, in order to contribute to the linguistic studies about the functioning of Brazilian Portuguese in this geographical context.

Who is in charge? An L2 Discourse Intonation Study on Four Prosodic Parameters to Exert the Pragmatic Function of Dominance and Control in the Context of L2 Non-specialist Public Speaking

This paper reports the findings from a study of the learning of English intonation by Spanish speakers within the discourse mode of L2 oral presentation. The purpose of this experiment is, firstly, to compare four prosodic parameters before and after an L2 discourse intonation training programme and, secondly, to confirm whether subjects, after the aforementioned L2 discourse intonation training, are able to match the form of these four prosodic parameters to the discourse-pragmatic function of dominance and control. The study designed the instructions and tasks to create the oral and written corpora and Brazil’s Pronunciation for Advanced Learners of English was adapted for the pedagogical aims of the present study. The learners’ pre- and post-tasks were acoustically analysed and a pre / post- questionnaire design was applied to interpret the acoustic analysis. Results indicate most of the subjects acquired a wider choice of the four prosodic parameters partly due to the prosodically-annotated transcripts that were developed throughout the L2 discourse intonation course. Conversely, qualitative and quantitative data reveal most subjects failed to match the forms to their appropriate pragmatic functions to express dominance and control in an L2 oral presentation.

Training Finnish and English Speakers to Produce Non-native Vowels

The subject of this thesis was the acquisition of difficult non-native vowels by speakers of two different languages. In order to study the subject, a group of Finnish speakers and another group of American English speakers were recruited and they underwent a short listen-and-repeat training that included as stimuli the semisynthetically created pseudowords /ty:ti/ and /tʉ:ti/. The aim was to study the effect of the training method on the subjects as well as the possible influence of the speakers’ native language on the process of acquisition. The selection of the target vowels /y/ and /ʉ/ was made according to the Speech Learning Model and Perceptual Assimilation Model, both of which predict that second language speech sounds that share similar features with sounds of a person’s native language are most difficult for the person to learn. The vowel /ʉ/ is similar to Finnish vowels as well as to vowels of English, whereas /y/ exists in Finnish but not in English, although it is similar to other English vowels. Therefore, it can be hypothesized that /ʉ/ is a difficult vowel for both groups to learn and /y/ is difficult for English speakers. The effect of training was tested with a pretest-training-posttest protocol in which the stimuli were played alternately and the subjects’ task was to repeat the heard stimuli. The training method was thought to improve the production of non-native sounds by engaging different feedback mechanisms, such as auditory and somatosensory. These, according to Template Theory, modify the production of speech by altering the motor commands from the internal speech system or the feedforward signal which translates the motoric commands into articulatory movements. The subjects’ productions during the test phases were recorded and an acoustic analysis was performed in which the formant values of the target vowels were extracted. Statistical analyses showed a statistically significant difference between groups in the first formant, signaling a possible effect of native motor commands. Furthermore, a statistically significant difference between groups was observed in the standard deviation of the formants in the production of /y/, showing the uniformity of native production. The training had no observable effect, possibly due to the short nature of the training protocol.

Método para reconhecimento de vogais e extração de parâmetros acústicos para analises forenses

Forensic speaker comparison exams have complex characteristics, demanding a long time for manual analysis. A method for automatic recognition of vowels, providing feature extraction for acoustic analysis is proposed, aiming to contribute as a support tool in these exams. The proposal is based in formant measurements by LPC (Linear Predictive Coding), selectively by fundamental frequency detection, zero crossing rate, bandwidth and continuity, with the clustering being done by the k-means method. Experiments using samples from three different databases have shown promising results, in which the regions corresponding to five of the Brasilian Portuguese vowels were successfully located, providing visualization of a speaker’s vocal tract behavior, as well as the detection of segments corresponding to target vowels.

Study of acoustic emission data analysis tools for structural health monitoring applications

Acoustic emission (AE) technique is one of the popular diagnostic techniques used for structural health monitoring of mechanical, aerospace and civil structures. But several challenges still exist in successful application of AE technique. This paper explores various tools for analysis of recorded AE data to address two primary challenges: discriminating spurious signals from genuine signals and devising ways to quantify damage levels.

Analysis of acoustic emission data for structural health monitoring applications

Acoustic emission (AE) is the phenomenon where high frequency stress waves are generated by rapid release of energy within a material by sources such as crack initiation or growth. AE technique involves recording these stress waves by means of sensors placed on the surface and subsequent analysis of the recorded signals to gather information such as the nature and location of the source. It is one of the several diagnostic techniques currently used for structural health monitoring (SHM) of civil infrastructure such as bridges. Some of its advantages include ability to provide continuous in-situ monitoring and high sensitivity to crack activity. But several challenges still exist. Due to high sampling rate required for data capture, large amount of data is generated during AE testing. This is further complicated by the presence of a number of spurious sources that can produce AE signals which can then mask desired signals. Hence, an effective data analysis strategy is needed to achieve source discrimination. This also becomes important for long term monitoring applications in order to avoid massive date overload. Analysis of frequency contents of recorded AE signals together with the use of pattern recognition algorithms are some of the advanced and promising data analysis approaches for source discrimination. This paper explores the use of various signal processing tools for analysis of experimental data, with an overall aim of finding an improved method for source identification and discrimination, with particular focus on monitoring of steel bridges.

Scaling acoustic data analysis through collaboration and automation

Monitoring and assessing environmental health is becoming increasingly important as human activity and climate change place greater pressure on global biodiversity. Acoustic sensors provide the ability to collect data passively, objectively and continuously across large areas for extended periods of time. While these factors make acoustic sensors attractive as autonomous data collectors, there are significant issues associated with large-scale data manipulation and analysis. We present our current research into techniques for analysing large volumes of acoustic data effectively and efficiently. We provide an overview of a novel online acoustic environmental workbench and discuss a number of approaches to scaling analysis of acoustic data; collaboration, manual, automatic and human-in-the loop analysis.

Characterizing the signal pattern of a four-cylinder diesel engine using acoustic emission and vibration analysis

Condition monitoring of diesel engines can prevent unpredicted engine failures and the associated consequence. This paper presents an experimental study of the signal characteristics of a 4-cylinder diesel engine under various loading conditions. Acoustic emission, vibration and in-cylinder pressure signals were employed to study the effectiveness of these techniques for condition monitoring and identifying symptoms of incipient failures. An event driven synchronous averaging technique was employed to average the quasi-periodic diesel engine signal in the time domain to eliminate or minimize the effect of engine speed and amplitude variations on the analysis of condition monitoring signal. It was shown that acoustic emission (AE) is a better technique than vibration method for condition monitor of diesel engines due to its ability to produce high quality signals (i.e., excellent signal to noise ratio) in a noisy diesel engine environment. It was found that the peak amplitude of AE RMS signals correlating to the impact-like combustion related events decreases in general due to a more stable mechanical process of the engine as the loading increases. A small shift in the exhaust valve closing time was observed as the engine load increases which indicates a prolong combustion process in the cylinder (to produce more power). On the contrary, peak amplitudes of the AE RMS attributing to fuel injection increase as the loading increases. This can be explained by the increase fuel friction caused by the increase volume flow rate during the injection. Multiple AE pulses during the combustion process were identified in the study, which were generated by the piston rocking motion and the interaction between the piston and the cylinder wall. The piston rocking motion is caused by the non-uniform pressure distribution acting on the piston head as a result of the non-linear combustion process of the engine. The rocking motion ceased when the pressure in the cylinder chamber stabilized.

Large scale participatory acoustic sensor data analysis : tools and reputation models to enhance effectiveness

Acoustic sensors play an important role in augmenting the traditional biodiversity monitoring activities carried out by ecologists and conservation biologists. With this ability however comes the burden of analysing large volumes of complex acoustic data. Given the complexity of acoustic sensor data, fully automated analysis for a wide range of species is still a significant challenge. This research investigates the use of citizen scientists to analyse large volumes of environmental acoustic data in order to identify bird species. Specifically, it investigates ways in which the efficiency of a user can be improved through the use of species identification tools and the use of reputation models to predict the accuracy of users with unidentified skill levels. Initial experimental results are reported.

Analysis of acoustic emission data for accurate damage assessment for structural health monitoring applications

Structural health monitoring (SHM) refers to the procedure used to assess the condition of structures so that their performance can be monitored and any damage can be detected early. Early detection of damage and appropriate retrofitting will aid in preventing failure of the structure and save money spent on maintenance or replacement and ensure the structure operates safely and efficiently during its whole intended life. Though visual inspection and other techniques such as vibration based ones are available for SHM of structures such as bridges, the use of acoustic emission (AE) technique is an attractive option and is increasing in use. AE waves are high frequency stress waves generated by rapid release of energy from localised sources within a material, such as crack initiation and growth. AE technique involves recording these waves by means of sensors attached on the surface and then analysing the signals to extract information about the nature of the source. High sensitivity to crack growth, ability to locate source, passive nature (no need to supply energy from outside, but energy from damage source itself is utilised) and possibility to perform real time monitoring (detecting crack as it occurs or grows) are some of the attractive features of AE technique. In spite of these advantages, challenges still exist in using AE technique for monitoring applications, especially in the area of analysis of recorded AE data, as large volumes of data are usually generated during monitoring. The need for effective data analysis can be linked with three main aims of monitoring: (a) accurately locating the source of damage; (b) identifying and discriminating signals from different sources of acoustic emission and (c) quantifying the level of damage of AE source for severity assessment. In AE technique, the location of the emission source is usually calculated using the times of arrival and velocities of the AE signals recorded by a number of sensors. But complications arise as AE waves can travel in a structure in a number of different modes that have different velocities and frequencies. Hence, to accurately locate a source it is necessary to identify the modes recorded by the sensors. This study has proposed and tested the use of time-frequency analysis tools such as short time Fourier transform to identify the modes and the use of the velocities of these modes to achieve very accurate results. Further, this study has explored the possibility of reducing the number of sensors needed for data capture by using the velocities of modes captured by a single sensor for source localization. A major problem in practical use of AE technique is the presence of sources of AE other than crack related, such as rubbing and impacts between different components of a structure. These spurious AE signals often mask the signals from the crack activity; hence discrimination of signals to identify the sources is very important. This work developed a model that uses different signal processing tools such as cross-correlation, magnitude squared coherence and energy distribution in different frequency bands as well as modal analysis (comparing amplitudes of identified modes) for accurately differentiating signals from different simulated AE sources. Quantification tools to assess the severity of the damage sources are highly desirable in practical applications. Though different damage quantification methods have been proposed in AE technique, not all have achieved universal approval or have been approved as suitable for all situations. The b-value analysis, which involves the study of distribution of amplitudes of AE signals, and its modified form (known as improved b-value analysis), was investigated for suitability for damage quantification purposes in ductile materials such as steel. This was found to give encouraging results for analysis of data from laboratory, thereby extending the possibility of its use for real life structures. By addressing these primary issues, it is believed that this thesis has helped improve the effectiveness of AE technique for structural health monitoring of civil infrastructures such as bridges.

Reputation modelling in citizen science for environmental acoustic data analysis

Citizen Science projects are initiatives in which members of the general public participate in scientific research projects and perform or manage research-related tasks such as data collection and/or data annotation. Citizen Science is technologically possible and scientifically significant. However, as the gathered information is from the crowd, the data quality is always hard to manage. There are many ways to manage data quality, and reputation management is one of the common approaches. In recent year, many research teams have deployed many audio or image sensors in natural environment in order to monitor the status of animals or plants. The collected data will be analysed by ecologists. However, as the amount of collected data is exceedingly huge and the number of ecologists is very limited, it is impossible for scientists to manually analyse all these data. The functions of existing automated tools to process the data are still very limited and the results are still not very accurate. Therefore, researchers have turned to recruiting general citizens who are interested in helping scientific research to do the pre-processing tasks such as species tagging. Although research teams can save time and money by recruiting general citizens to volunteer their time and skills to help data analysis, the reliability of contributed data varies a lot. Therefore, this research aims to investigate techniques to enhance the reliability of data contributed by general citizens in scientific research projects especially for acoustic sensing projects. In particular, we aim to investigate how to use reputation management to enhance data reliability. Reputation systems have been used to solve the uncertainty and improve data quality in many marketing and E-Commerce domains. The commercial organizations which have chosen to embrace the reputation management and implement the technology have gained many benefits. Data quality issues are significant to the domain of Citizen Science due to the quantity and diversity of people and devices involved. However, research on reputation management in this area is relatively new. We therefore start our investigation by examining existing reputation systems in different domains. Then we design novel reputation management approaches for Citizen Science projects to categorise participants and data. We have investigated some critical elements which may influence data reliability in Citizen Science projects. These elements include personal information such as location and education and performance information such as the ability to recognise certain bird calls. The designed reputation framework is evaluated by a series of experiments involving many participants for collecting and interpreting data, in particular, environmental acoustic data. Our research in exploring the advantages of reputation management in Citizen Science (or crowdsourcing in general) will help increase awareness among organizations that are unacquainted with its potential benefits.

Reputation modelling in Citizen Science for environmental acoustic data analysis

Citizen Science projects are initiatives in which members of the general public participate in scientific research projects and perform or manage research-related tasks such as data collection and/or data annotation. Citizen Science is technologically possible and scientifically significant. However, although research teams can save time and money by recruiting general citizens to volunteer their time and skills to help data analysis, the reliability of contributed data varies a lot. Data reliability issues are significant to the domain of Citizen Science due to the quantity and diversity of people and devices involved. Participants may submit low quality, misleading, inaccurate, or even malicious data. Therefore, finding a way to improve the data reliability has become an urgent demand. This study aims to investigate techniques to enhance the reliability of data contributed by general citizens in scientific research projects especially for acoustic sensing projects. In particular, we propose to design a reputation framework to enhance data reliability and also investigate some critical elements that should be aware of during developing and designing new reputation systems.

Practical analysis of big acoustic sensor data for environmental monitoring

Monitoring the environment with acoustic sensors is an effective method for understanding changes in ecosystems. Through extensive monitoring, large-scale, ecologically relevant, datasets can be produced that can inform environmental policy. The collection of acoustic sensor data is a solved problem; the current challenge is the management and analysis of raw audio data to produce useful datasets for ecologists. This paper presents the applied research we use to analyze big acoustic datasets. Its core contribution is the presentation of practical large-scale acoustic data analysis methodologies. We describe details of the data workflows we use to provide both citizen scientists and researchers practical access to large volumes of ecoacoustic data. Finally, we propose a work in progress large-scale architecture for analysis driven by a hybrid cloud-and-local production-grade website.

Acoustic identification of 12 species of echolocating bat by discriminant function analysis and artificial neural networks

We recorded echolocation calls from 14 sympatric species of bat in Britain. Once digitised, one temporal and four spectral features were measured from each call. The frequency-time course of each call was approximated by fitting eight mathematical functions, and the goodness of fit, represented by the mean-squared error, was calculated. Measurements were taken using an automated process that extracted a single call from background noise and measured all variables without intervention. Two species of Rhinolophus were easily identified from call duration and spectral measurements. For the remaining 12 species, discriminant function analysis and multilayer back-propagation perceptrons were used to classify calls to species level. Analyses were carried out with and without the inclusion of curve-fitting data to evaluate its usefulness in distinguishing among species. Discriminant function analysis achieved an overall correct classification rate of 79% with curve-fitting data included, while an artificial neural network achieved 87%. The removal of curve-fitting data improved the performance of the discriminant function analysis by 2 %, while the performance of a perceptron decreased by 2 %. However, an increase in correct identification rates when curve-fitting information was included was not found for all species. The use of a hierarchical classification system, whereby calls were first classified to genus level and then to species level, had little effect on correct classification rates by discriminant function analysis but did improve rates achieved by perceptrons. This is the first published study to use artificial neural networks to classify the echolocation calls of bats to species level. Our findings are discussed in terms of recent advances in recording and analysis technologies, and are related to factors causing convergence and divergence of echolocation call design in bats.