Search

There is a general consensus among scholars that one same sentence can be said in different ways. This consensus, however, falls apart when the issues are discussed related to the efficiency of prosody in allowing us to say the very same sequence of segments with different meanings. According to Hirst (1987) e t’Hartet. al. (1990) the difficulties of transcribing prosody overcome the difficulties of transcribing the segmental. They argue that in order to investigate how the melody contributes to the overall understanding of an enunciation a metalanguage was required through which the phenomenon can be discussed. The vocabulary of this metalanguage must consist of appropriate descriptive units through which it is possible to speak of entities and structures at various levels of observation. The semi-automatic analysis, by means of the programs MOMEL and INTSINT, was used in the description of some intonational languages such as: French (di Cristo, 2011), British English (Auran, Bouzon and Hirst, 2004; Brierley, 2011) and analysis of the Brazilian Portuguese rhythm by means of manual INTSINT (Gonçalves, 2000). The present study aims to determine if the program of semi-automatic intonational analysis - INTSINT - is capable of reproduce melodic variation trends of the Brazilian Portuguese. In our proposal, the effectiveness of the program to differentiate between declarative and interrogative forms during reading was evaluated. The following hypotheses were raised: (I) the analysis of intonation with the program INTSINT allows the choice of different phonological theories regarding the use of entonative units;(II) it is possible to distinguish between declarative and interrogative forms by means of coded target points to Brazilian Portuguese. An explanatory analysis of the codification performed by MOMELINTSINT was performed. The corpus of this study consists of 10 short texts, with approximately five sentences that were read by 10 people of the female sex, aged 20-30 years. The acoustic analysis of the data was performed using the program Praatversion 4.4.27, available at www.praat.org (Weenink and Boersma, 1997 ,i contains the necessary extensions for the implementation of the programs MOMEL / INTSINT. These last ones are freely available on the site http://aune.lpl.univaix.fr/~auran/english/ressources.html. The high beginning was common for the declarative and interrogative modalities. This fact was represented by the first target point, coded as T, or the second target point, coded as M; but followed by H, T or U. Nevertheless, they differ at the end of the unit: in interrogative sentences, a sharp increase represented by the symbols T and H can be observed just before the final fall. In declarative sentences it is possible to perceive a rise is followed by a downfall, but not as much and not as systematically as in the interrogative sentences. Far from closing the discussion on intonation units and in the chosen modalities for this study,our initial goal of evaluating the behavior of the program INTSINT against pitch variations was reached. We conclude that the program INTSINT is able to transmit trends and also the general melodic pattern of F0 curve.

International audience; This article describes an experiment in paratone detection based on a spoken corpus of English for Academic Purposes (EAP) recently automatically re-annotated with prosodic information. The Momel and INTSINT annotations were carried out using SPPAS. The EIIDA corpus was chosen as it offered long uninterrupted stretches of speech of academic presentations. We describe the clustering method adopted for automatic detection, contrasting a supervised and an unsupervised method of paratone boundary detection. We showcase the relevance of the annotation scheme followed for this corpus and contribute to the investigation of the phonostyle of lecture delivery. We discuss the relevance of clustering methods applied to the labels of the pitch targets for the analysis of paratones.

This paper presents a multilingual learners corpus, AixOx, collect-ed in the framework of an Alliance project (a partnership between the British Council and The French Ministry of Foreign Affairs). The corpus consists of the recording of 40 1-minute passages in English and French from the Eurom 1 corpus (Chan et al., 1995), read by native speakers and L2 learners. French native speakers reading the French and English passages were recorded in Aix-en-Provence, and English native speakers reading the English and French passages were recorded in Oxford. The AixOx corpus con-tains about 40 hours of read speech and can be downloaded from the “Speech and Language Data Repository” (http://sldr.org). This paper also presents the tools used for automatic anno-tation on several layers using algorithms: •SPPAS –SPeech Phonetization Alignment and Syllabifica-tion– (Bigi, 2012) for a segmentation into utterances, words, syllables and phonemes;•MoMel –Modelling Melody– and INTSINT –INternational Transcription System for INTonation– (Hirst, 2007) for the modelling and coding of intonation.Finally, an example of a pedagogical application of the cor-pus is given: a pilot-study on the intonation of questions. We show how the AixOx corpus can be used to compare the produc-tions of natives with learners and how it is possible, thanks to the annotation, to understand the prosodic realisations (whether they be positive or negative) and explain them. We conclude that AixOx, with its multi-layered annotation, is a very rich oral data-base for all kinds of studies on L1 productions, L2 productions, language contact, both at the segmental and supra-segmental levels since it offers a phonemic segmentation and alignment and a pro-sodic labelling.

International audience; This paper describes the application of the analysis by synthesis paradigm to the melody of speech. A complete chain of processes is described from the acoustic analysis of fundamental frequency (f0), via the phonetic modelling of f0 using the Momel algorithm, to the surface phonological representation of the curves using the INTSINT alphabet. Each step of the chain is designed as a reversible process which can be used to generate an acoustic output allowing an objective evaluation of the analysis. Finally, the current implementation of ProZed, a prosody editor for linguists, is described. It is argued that an explicit set of modelling tools like this will allow linguists to test different models of phonological structure which, it is hoped, will result in the availability of more and better data on a wide variety of languages.

En este trabajo se evalúa empíricamente el reconocimiento automático de actos de diálogo. Se usan datos provenientes de un corpus de diálogos con habla espontánea. En cada diálogo dos hablantes colaboran en el diseño de cocinas usando herramientas C.A.D.; uno de ellos desempeña el rol del Sistema y el otro el del Usuario. Los actos de diálogo se etiquetan con DIME-DAMSL, esquema que considera dos planos de expresión: obligaciones y common ground. La evaluación se realiza probando modelos clasificadores creados con algoritmos de aprendizaje máquina: uno para obligaciones y otro para common ground. El principal dato predictor analizado es el acto de diálogo correspondiente al enunciado inmediato anterior. Se pondera también la contribución de información adicional, como la entonación, etiquetada con INTSINT, la modalidad del enunciado, el rol del hablante y el tipo de acto de diálogo del plano complementario. Una aplicación práctica sería en sistemas de administración de diálogo. In this paper the automatic recognition of dialogue acts is evaluated on an empirical basis. Data from a dialogue corpus with spontaneous speech are used. In each dialogue two speakers collaborate to design a kitchen using a C.A.D. software tool; one of them plays the System’s role and the other plays the User’s role. Dialogue acts are annotated with DIMEDA MSL, a scheme considering two expression planes: obligations and common ground. The evaluation is performed by testing classification models created with Machine Learning algorithms: one model for obligations and other for common ground. The mainly analyzed predictor data is the dialogue act corresponding to the immediately previous utterance. The contribution of other information sources is also evaluated, such as intonation, annotated with INTSINT, utterance mood, speaker role and dialogue act type of the complementary expression plane. A practical application can be the implementation of dialogue management systems.

Synthesis of adequate prosody from text is considered as one of the crucial issues in order to achieve high quality speech synthesis. This problem has not been addressed yet in our speech synthesis system. In this paper we present our first attempt to model intonation, with selected tools - MOMEL and INTSINT. They present low-level coding of F0 contour and more abstract description of speech melody respectively. Finally, we present our experiments in coding INTSINT symbols from MOMEL targets and our results in re-synthesizing the targets from the symbols. The results are comparable to those in other studies regarding different languages.

International audience; The way in which prosody contributes to meaning is still, today, a poorly understood process corresponding to a mapping between two levels of representation, for neither of which there is any general consensus. It is argued that annotation of prosody generally consists in describing both prosodic function and prosodic form, but that it would be preferable to clearly distinguish the two levels. One elementary annotation system for prosodic function, IF-annotation, is, it has been argued, sufficient to capture at least those aspects of prosodic function which influence syntactic interpretation. The annotation of prosodic form can be carried out automatically by means of an F0 modelling algorithm, MOMEL, and an automatic coding scheme, INTSINT. The resulting annotation is underdetermined by the IF-annotation, but defining mapping rules between representations of function and representation of form could provide an interesting means of establishing an enriched functional annotation system through analysis by synthesis.