Un modèle de la facilité d’écoute des documents audios pour les apprenants du français langue étrangère
(2025)
Files
Thèse_Minami_Ozawa.pdf
Open Access - Adobe PDF
- 2.81 MB
Details
- Authors
- Supervisors
- Abstract
- This thesis proposes an automatic prediction model for assessing the listenability of French audio documents. The aim of this model is to support the teaching and learning of oral comprehension among French language learners by facilitating the calibration of the difficulty level of the documents used in training. To develop this model, we first collected a corpus of audio documents and their transcriptions from 25 French as a Foreign Language textbooks. We then identified a set of variables to integrate into the model: linguistic variables based on the transcription and acoustic variables derived from the audio. The data was classified according to speech styles—dialogue and monologue—and specific models were developed for each. The analysis conducted in this thesis regarding predictive models of listenability is divided into two main stages. A first model was created based on manually corrected data. Then, a second model was automatically created without any manual data correction. By comparing the results of these two models, we measured the performance of an automatic predictive model for listenability. For each stage of analysis, three prediction approaches were considered: SVM, wav2vec, and CamemBERT. As a result, it was found that a model using linguistic variables from the transcription was effective in predicting listenability. Regarding fluency variables obtained from the audio, their correlation with listenability was higher in dialogues. Furthermore, model analysis revealed that these fluency variables derived from audio contributed to the improvement of model performance, even though they remained inferior to linguistic variables. During the automation of the model, it was identified that automatic speech alignment represented a weakness. Considering this limitation, it is necessary to further examine acoustic variables and explore the implementation of an automatic predictive model for listenability.
- Affiliations
Citations
Ozawa, M. (2025). Un modèle de la facilité d’écoute des documents audios pour les apprenants du français langue étrangère. https://hdl.handle.net/2078.5/260731