Electrical and Computers Eng. Department

This volume presents the 14th edition of the International Conference on the Computational Processing of Portuguese (PROPOR 2020). The conference was held at the Colégio do Espírito Santo, Universidade de Évora, in Évora, Portugal, during March 2-4, 2020. Although the conference is in its 14th edition, it is celebrating the maturity of 27 years of existence, since it has been predominantly a biannual event. In that route it returns to Évora after 21 years. PROPOR is the main scientific meeting in the area of language and speech technologies for Portuguese and on the basic and applied research issues related to this language. The meeting is a very rich forum for researchers dedicated to the computational processing of Portuguese, promoting the exchange of experiences in the development of methodologies, language resources, tools, applications, and innovative projects. It is also a meeting for the reception of new members to the community, represented mainly by the students in the field. A total of 70 submissions were received for the main event, involving 233 authors from many countries worldwide, such as Portugal,

This paper addresses the problem of grapheme to phoneme conversion in order to create a pronunciation dictionary from a vocabulary of the most frequent words in European Portuguese. A system based on a mixed approach funded on a stochastic model with embedded rules for stressed vowel assignment is described. The model can generate pronunciations from unrestricted words; however, a dictionary with the 40k most frequent words was constructed and corrected interactively. The vocabulary was defined using the CETEMPúblico corpus. The model and dictionary are publicly available.

Spontaneous speech is full of hesitations, such as fillers, word cutoffs , repetitions and segmental extensions. Automatic identification of such hesitations has several applications; however, it is a challenging research problem. In this paper acoustic-phonetic properties of hesitation phenomena are explored in order to identify and annotate some of these events in a spontaneous speech corpus of Portuguese broadcast television news. Based on pitch, energy, spectral and durational characteristics of the filled pauses and segmental extensions during their production, we intend to characterize the acoustic-phonetic regularity of the phenomena. A speech recognition system was used to help locating the filled pauses and extensions. The events detected were then manually validated. Our preliminary results suggest that there are regular trends in the production of these hesitation events, which could distinguish them from other events within the structure of Portuguese. Our purpose with this work is to improve acoustic modeling for spontaneous speech recognition systems. Some insights into the process of human speech communication for Portuguese are gained as well.

This paper presents research results on new syllable structures for the Portuguese language. The location of the syllable boundaries is a well known problem of nonconsensual resolution in Portuguese, mainly when the acoustic-phonetic constraints are taken into account. Based on various acoustic-phonetic restrictions of Portuguese speech, new syllable splitting rules were studied in a corpus of 400K Portuguese words. All these words were automatically converted into sequences of C (consonants) and V (vowels) and also related to the new syllable structures. When syllable structures were outlined in a statistical form, some mapping confirmed our expectations, others apparently ran against the expectations and others appeared to be new. These results provide clues about Portuguese syllabification architecture and can also guide the development of improved complementary models for syllable processing in speech applications.

This paper addresses the problem of grapheme to phoneme conversion in order to create a pronunciation dictionary from a vocabulary of the most frequent words in European Portuguese. A system based on a mixed approach funded on a stochastic model with embedded rules for stressed vowel assignment is described. The model can generate pronunciations from unrestricted words; however, a dictionary with the 40k most frequent words was constructed and corrected interactively. The vocabulary was defined using the CETEMPúblico corpus. The model and dictionary are publicly available.

This study presents an approach to the task of automatically classifying and detecting speaking styles. The detection of speaking styles is useful for the segmentation of multimedia data into consistent parts and has important applications, such as identifying speech segments to train acoustic models for speech recognition. In this work the database consists of daily news broadcasts in Portuguese television, on which two main speaking styles are evident: read speech from voice-over and anchors, and spontaneous speech from interviews and commentaries. Using a combination of phonetic and prosodic features we can separate these two speaking styles with a good accuracy (93.7% read, 69.5% spontaneous). This is performed in two steps. The first step separates the speech segments from the non-speech audio segments and the second step classifies read versus spontaneous speaking style. The use of phonetic and prosodic features provides alternative information that leads to an improvement of the classification and detection task.

This paper presents a study of European Portuguese elderly speech, in which the acoustic characteristics of two groups of elderly speakers (aged 60-75 and over 75) are compared with those of young adult speakers (aged 19-30). The correlation between age and a set of 14 acoustic features was investigated, and decision trees were used to establish the relative importance of the features. A greater use of pauses characterized speakers aged 60 and over. For female speakers, speech rate also appeared to correlate with age. For male speakers, jitter distinguished between speakers aged 60-75 and older. The correlation between the features and speech recognition performance was also investigated. Word error rate correlated mostly with the use of pauses, speech rate, and the ratio of long phone realizations. Finally, by comparing the phone sequences used by the recognizer on the most frequent words, we observed that the young adult speakers reduced schwas more than the elderly speakers. This result seems to confirm the common idea that young speakers reduce articulation more than older speakers. Further investigation is needed to confirm this result by determining whether this is due to ageing or to the generation gap.

This paper describes a project with the overall goal of providing a natural interaction communication platform accessible and adapted for all users, especially for people with speech impairments and elderly, by sharing knowledge between Industry and Academia. The platform will adopt the principles of natural user interfaces such as speech, silent speech, gestures, pictograms, among others, and will provide a set of services that allow easy access to social networks, friends and remote family members, thus contributing to overcome social-exclusion of people with special needs or impairments. Application of these features will be performed in the context of serious games, virtual reality environments and assisted living scenarios. The project will be executed in the scope of the Marie Curie Action Industry-Academia Partnerships and Pathways and will bring together the knowledge of five partners, from three different countries, Portugal, Spain and Turkey. This synergy will be based on transfer of knowledge mechanisms such as regular assignments and seminars, and is expected to have repercussions in scientific, technological, social and economic domains.

Automatically recognising children's speech is a very difficult task. This difficulty can be attributed to the high variability in children's speech, both within and across speakers. The variability is due to developmental changes in children's anatomy, speech production skills et cetera, and manifests itself, for example, in fundamental and formant frequencies, the frequency of disfluencies, and pronunciation quality. In this paper, we report the results of acoustic and auditory analyses of 3-10-year-old European Portuguese children's speech. Furthermore, we are able to correlate some of the pronunciation error patterns revealed by our analyses-such as the truncation of consonant clusters-with the errors made by a children's speech recogniser trained on speech collected from the same age group. Other pronunciation error patterns seem to have little or no impact on speech recognition performance. In future work, we will attempt to use our findings to improve the performance of our recogniser.

A conversão de grafema para fonema diz respeito à tarefa de encontrar a pronúncia de um vocábulo dado na sua forma escrita, a qual tem uma forte componente de aplicação em sistemas de reconhecimento e de síntese de fala. Uma nova abordagem na conversão de grafema para fonema é proposta, aplicando um modelo híbrido para o qual concorrem regras fonológicas e decisões estatísticas. Os resultados mostram que a incorporação de regras fonológicas em algoritmos de informação estatística melhora acentuadamente o desempenho do conversor. Para este trabalho, foi construído um dicionário de pronunciação com mais de 40000 vocábulos derivados do corpus CETEMPúblico. Os dicionários fonológicos de pronunciação para o português europeu, bem como outros recursos produzidos durante este trabalho, estão disponibilizados publicamente. O sistema que aqui se descreve foi aplicado à língua portuguesa escrita, sem e com o Acordo Ortográfico de 1990, e, ainda que aplicado ao português na sua vertente europeia, observa características que permitem a sua aplicação a outras línguas românicas.

One of the frequent symptoms of Parkinson's disease (PD) is a progressive dysarthria, the difficulty with articulation. Characterization of dysarthric speech is important, for instance, for the adaptation of speech recognition systems to PD patients. In this work, we investigate acoustic and phonetic-prosodic characteristics of speech produced by PD patients in the European Portuguese. For this purpose, a speech database, consisting mainly on phonetically rich sentences, has been collected from a group of patients with different degrees of PD severity and a healthy control group. Only vowels in a stressed position in continuous speech context (i.e. no sustained vowels) were analyzed. Results show a centralization of vowel formant frequencies for PD speech, as expected. However, some of the usual features for discriminating PD speech were not always found to be statistically significant. Furthermore, an analysis of acoustic, spectral and prosodic features towards classifying PD speech shows that dynamic features are of highest importance in this task.

In this paper we present a free, open source platform, that translates in real time (written) European Portuguese into Portuguese Sign Language, being the signs produced by an avatar. We discuss basic needs of such a system in terms of Natural Language Processing and Animation Synthesis, and propose an architecture for it. Moreover, we have selected a set of existing tools that couple with our free, open-source philosophy, and implemented a prototype with them. Several case studies were conducted. A preliminary evaluation was done and, although the translation possibilities are still scarce and some adjustments still need to be done, our platform was already much welcomed by the deaf community.

Several efforts have been done towards the development of platforms that allow the translation of specific sign languages to the correspondent spoken language (and vice-versa). In this (demo) paper, we describe a freely available system that translates, in real time, European Portuguese (text) into Portuguese Sign Language (LGP), by using an avatar. We focus in how some linguistic elements are dealt in LGP, and in the Natural Language Processing (NLP) step implemented in our system. The system's interface will be used to demonstrate it. Although only a small set of linguistic phenomena are implemented, it can be seen how the system copes with it.

Cumpre-me manifestar o meu reconhecimento a todos quanto contribuíram para a concretização deste trabalho. Presto o meu penhorado tributo à opinião avisada do Prof. Doutor Telmo dos Santos Verdelho, com quem pude contar, ao longo deste trabalho, com a ajuda solícita. Ao senhor Prof. Doutor Jorge Morais Barbosa, pelo generoso acolhimento científico e humano que sempre me proporcionou, neste momento como em anteriores, facilitando-me o acesso à sua biblioteca e ao seu saber, nas diferentes ocasiões em que se disponibilizou a comigo discutir diversos aspectos da presente dissertação. O trabalho de investigação nem sempre se revela corredio e alentador, mas, em todos os momentos, encontrei no Professor o orientador dedicado, esclarecido, sensível e paciente. Ao mestre e amigo muito agradeço. Uma palavra de agradecimento é igualmente devida a meus Mestres da Faculdade de Letras da Universidade de Coimbra, de quem recebi, ao longo da minha formação universitária, lições de humanismo e de sabedoria inestimáveis. Aos Colegas linguistas o meu obrigada pelo encorajamento e disponibilidade. Aos meus Amigos um agradecimento muito especial à sensibilidade com a qual tiveram a bondade de me incentivar. Aos meus Pais, em tudo inexcedíveis, ao meu Marido, pelo envolvimento profundamente solidário com que acompanhou a feitura desta tese, não há palavras ou dedicatórias que traduzam a minha gratidão. Ao Henrique e ao Rafael, meus filhos, a quem também dedico este trabalho, porque se privaram, com elevada compreensão, de uma mãe mais presente. 14 A título de enquadramento, sabe-se que este trabalho se faz acompanhar de quatro mapas, sendo os números dois e três uma reprodução algo retocada e melhorada esteticamente dos mapas de 1897 e de 1929, elaborados por Leite de Vasconcellos.

To evaluate the reading performance of children, human assessment is usually involved, where a teacher or tutor has to take time to individually estimate the performance in terms of fluency (speed, accuracy and expression). Automatic estimation of reading ability can be an important alternative or complement to the usual methods, and can improve other applications such as elearning. Techniques must be developed to analyse audio recordings of read utterances by children and detect the deviations from the intended correct reading i.e. disfluencies. For that goal, a database of 284 European Portuguese children from 6 to 10 years old (1st-4th grades) reading aloud amounting to 20 hours was collected in private and public Portuguese schools. This paper describes the design of the reading tasks as well as the data collection procedure. The presence of different types of disfluencies is analysed as well as reading performance compared to known curricular goals.

Reading aloud performance in children is typically assessed by teachers on an individual basis, manually marking reading time and incorrectly read words. A computational tool that assists with recording reading tasks, automatically analyzing them and providing performance metrics could be a significant help. Towards that goal, this work presents an approach to automatically predicting the overall reading aloud ability of primary school children (6-10 years old), based on the reading of sentences and pseudowords. The opinions of primary school teachers were gathered as ground truth of performance, who provided 0-5 scores closely related to the expectations at the end of each grade. To predict these scores automatically, features based on reading speed and number of disfluencies were extracted, after an automatic disfluency detection. Various regression models were trained, with Gaussian process regression giving best results for automatic features. Feature selection from both sentence and pseudoword reading tasks gave the closest predictions, with a correlation of 0.944. Compared to the use of manual annotation with the best correlation being 0.952, automatic annotation was only 0.8% worse. Furthermore, the error rate of predicted scores relative to ground truth was found to be smaller than the deviation of evaluators' opinion per child.

To automatically evaluate the performance of children reading aloud or to follow a child's reading in reading tutor applications, different types of reading disfluencies and mispronunciations must be accounted for. In this work, we aim to detect most of these disfluencies in sentence and pseudoword reading. Detecting incorrectly pronounced words, and quantifying the quality of word pronunciations, is arguably the hardest task. We approach the challenge as a two-step process. First, a segmentation using task-specific lattices is performed, while detecting repetitions and false starts and providing candidate segments for words. Then, candidates are classified as mispronounced or not, using multiple features derived from likelihood ratios based on phone decoding and forced alignment, as well as additional meta-information about the word. Several classifiers were explored (linear fit, neural networks, support vector machines) and trained after a feature selection stage to avoid overfitting. Improved results are obtained using feature combination compared to using only the log likelihood ratio of the reference word (22% versus 27% miss rate at constant 5% false alarm rate).

Evaluating children's reading aloud proficiency is typically a task done by teachers on an individual basis, where reading time and wrong words are marked manually. A computational tool that assists with recording reading tasks, automatically analyzing them and outputting performance related metrics could be a significant help to teachers. Working towards that goal, this work presents an approach to automatically predict the overall reading aloud ability of primary school children by employing automatic speech processing methods. Reading tasks were designed focused on sentences and pseudowords, so as to obtain complementary information from the two distinct assignments. A dataset was collected with recordings of 284 children aged 6-10 years reading in native European Portuguese. The most common disfluencies identified include intra-word pauses, phonetic extensions, false starts, repetitions, and mispronunciations. To automatically detect reading disfluencies, we first target extra events by employing task-specific lattices for decoding that allow syllable-based false starts as well as repetitions of words and sequences of words. Then, mispronunciations are detected based on the log likelihood ratio between the recognized and target words. The opinions of primary school teachers were gathered as ground truth of overall reading aloud performance, who provided 0-5 scores closely related to the expected performance at the end of each grade. To predict these scores, various features were extracted by automatic annotation and regression models were trained. Gaussian process regression proved to be the most successful approach. Feature selection from both sentence and pseudoword tasks give the closest predictions, with a correlation of 0.944 compared to the teachers' grading. Compared to the use of manual annotation, where the best models obtained give a correlation of 0.949, there was a relative decrease of only 0.5% for using automatic annotations to extract features. The error rate of predicted scores relative to ground truth also proved to be smaller than the deviation of evaluators' opinion per child.

This paper proposes an approach to automatically parse children's reading of sentences by detecting word pronunciations and extra content, and to classify words as correctly or incorrectly pronounced. This approach can be directly helpful for automatic assessment of reading level or for automatic reading tutors, where a correct reading must be identified. We propose a first segmentation stage to locate candidate word pronunciations based on allowing repetitions and false starts of a word's syllables. A decoding grammar based solely on syllables allows silence to appear during a word pronunciation. At a second stage, word candidates are classified as mispronounced or not. The feature that best classifies mispronunciations is found to be the log-likelihood ratio between a free phone loop and a word spotting model in the very close vicinity of the candidate segmentation. Additional features are combined in multifeature models to further improve classification, including: normalizations of the log-likelihood ratio, derivations from phone likelihoods, and Levenshtein distances between the correct pronunciation and recognized phonemes through two phoneme recognition approaches. Results show that most extra events were detected (close to 2% word error rate achieved) and that using automatic segmentation for mispronunciation classification approaches the performance of manual segmentation. Although the log-likelihood ratio from a spotting approach is already a good metric to classify word pronunciations, the combination of additional features provides a relative reduction of the miss rate of 18% (from 34.03% to 27.79% using manual segmentation and from 35.58% to 29.35% using automatic segmentation, at constant 5% false alarm rate).