Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/136086
Title: Aplicación de métodos de aprendizaje semi-supervisados para el reconocimiento del habla en personas con afasia
Author: Romero Ferrón, Mónica
Tutor: González Torre, Iván
Others: Conesa, Jordi  
Abstract: Traditionally, automatic speech recognition (ASR) systems require algorithms that use labeled databases for learning. However, a recent novel approach develops semi-supervised models that have the ability to perform part of their training on unlabeled data, thus facilitating their use in environments where labeled data is scarce. This research work is focused on the application of these learning methods in the health domain and, more specifically, on pathological voices coming from speakers with different types of aphasia. We have worked with the reference database AphasiaBank, which contains 78 hours of audios from patients with different degrees of aphasia, and which has already been used by other research groups. At the modeling level, the semi-supervised learning architecture used on this domain data has been optimized and tuned through the application of the Grid Search technique and the exhaustive search of the hyperparameters of the model. In this study, the results obtained are compared with those reflected in the state of the art. It is shown that the obtained recognition model presents results that improve other types of previously published approaches.
Keywords: NLP
automatic speech recognition
RAH
wav2vec2.0
aphasia
aphasia
neural networks
NLP
Document type: info:eu-repo/semantics/masterThesis
Issue Date: 6-Jun-2021
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
moniromeTFM0621memoria.pdfMemòria del TFM1,41 MBAdobe PDFThumbnail
View/Open