Internship title: Prosodic salience in oral corpora: the contribution of machine learning and signal processing for their identification

Context & objective and its expected results:

===============================

The highlighting of certain elements of an utterance determines the semantic and pragmatic interpretation in the corresponding sentence.

This emphasis is objectified by attention markers (salience), which in the case of oral corpora is limited to a few prosodic features [1].

The originality of the subject is to explore the contribution of signal processing and other communities working on salience [2] to identify new prosodic features useful for the identification of prosodic salience occurrences in oral corpora, in association with current machine learning approaches.

Previous works have enabled the development of expertise on the prosodic characterisation of utterances for the purposes of automatic conviction detection or automatic classification of conviction [3, 4, 5] or injunction [6].

Internship Objectives.

=============

The first task is a bibliographical study of the features used in linguistics for prosodic salience, as well as on salience markers used in other fields.

A second task will be the development of an automatic system for identifying prosodic salience occurrences, by learning a set of the occurrences identified by expert linguists.

This system could be hybrid, with or without deep learning, with the objective of evaluating the relevance of the markers.

The third task will be the confrontation of this system with a wide large oral corpus, as well as with a corpus being created for reading to children.

Required profile

===============

The candidate should have a background in machine learning and signal processing with an interest in audio speech processing.

A taste for programming and data processing is appreciated.

The application must include

– a CV

– a letter of motivation

– transcripts of grades from M1 (or equivalent level) and if possible M2

* Working place : PRISME Laboratory, University of Orleans, France

* Gratuity about 550 euros / month

* Duration : 5 up 6 months (suitable start: March 2022)

* Contacts : Philippe.ravier@univ-orleans.fr

Bibliography

[1] R. GODEMENT-BERLINE, Contribution à l'étude de la focalisation prosodique en français, Actes de la conférence conjointe JEP-TALN-RECITAL, pp. 164-172, 2016.

[2] F. LANDRAGIN, De la saillance visuelle à la saillance linguistique, Saillance. Aspects linguistiques et communicatifs de la mise en évidence dans un texte, Presses Universitaires de Franche-Comté, pp.67-84, 2011, Annales Littéraires de l’Université de Franche-Comté.

[3] A. HACINE-GHARBI, M. PETIT, P. RAVIER, F. NEMO, Prosody Based Automatic Classification of the Uses of French ‘oui’ as Convinced or Unconvinced Uses, 4th International Conference on Pattern Recognition Applications and Methods, ICPRAM, pp. 349- 354, Lisbonne, Portugal, 10-12 January, 2015.

[4] A. HACINE-GHARBI, P. RAVIER, F. NEMO, Local and Global Feature Selection for Prosodic Classification of the Word’s Uses, 6th International Conference on Pattern Recognition Applications and Methods, ICPRAM, pp. 711- 717, Porto, Portugal, February 2017.

[5] A. HACINE-GHARBI, P. RAVIER, Automatic classification of French spontaneous oral speech into injunction and no-injunction classes, 9th International Conference on Pattern Recognition Applications and Methods, ICPRAM, La Valetta, Malta, February 2020.

[6] A. BOUGRINE, P. RAVIER, A. HACINE-GHARBI, H. OUACHOUR LSTM Network based on Prosodic Features for the Classification of Injunction in French Oral Utterances, 11th International Conference on Pattern Recognition Applications and Methods, ICPRAM, Vienna, Austria, (virtual) February 2022