Inference of pronunciation difficulty from non-native data

01/28/2021

ELSA Corp. developed a system that assists its users to improve their American English accent. In order to develop exercises, appropriate for the level of its users, it is important to have a metric capable of assessing the difficulty of their exercises, according to the user’s proficiency level. Therefore, the objective of this thesis is to develop a system capable of determining the pronunciation difficulty associated with a certain utterance and its phonemes, for a Vietnamese student of English. Our model uses a Neural Network in order to forecast the probabilities associated to how competently the user pronounce each of the utterance’s phonemes. Then, using these probabilities, the system computes the difficulty score associated to the phoneme and the difficulty score associated to the utterance. In the end, we have a system able to receive as input an utterance and the proficiency level of the user. Then, the system outputs difficulty scores for the utterance and its phonemes.

Authors in the community:

• 

  J

  João Pedro de Sousa Correia

  ist425397

Supervisors of this institution:

• 

  I

  Isabel Maria Martins Trancoso

  ist11803

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico