Ir directamente a la navegación principal Ir directamente a la búsqueda Ir directamente al contenido principal

Increasing the Precision of Dysarthric Speech Intelligibility and Severity Level Estimate

Producción científica: Conference contribution › revisión exhaustiva

7 Citas (Scopus)

Resumen

Dysarthria is a speech disorder often characterized by slow speech with reduced intelligibility. Automated assessment of the severity-level and intelligibility of dysarthric speech can improve the efficiency and reliability of clinical assessment as well as benefit automatic speech recognition systems (ASR). However, in order to evaluate them, there are not sentence-level severity and intelligibility label. We only have access to speaker-per-level severity and intelligibility labels. This is a problem as dysarthric talkers might be able to produce some intelligible utterances due to frequent use and short utterances. Therefore, label based analysis might not be very accurate. To address this problem, we explore methods to estimate the severity-level and speech intelligibility in dysarthria given discrete speaker-level labeling in the training set. To accomplish this, we propose a machine learning based method using one-dimensional Convolutional Neural Networks (1-D CNN). The TORGO dataset is used to test the performance of the proposed method, with the UASpeech dataset used for Transfer learning (TL). To evaluate, an Averaged Ranking Score (ARS) and intelligibility probability distribution are used. Our findings demonstrate that the proposed method can assess speakers based on severity-level and intelligibility to provide a more granular analysis of factors underlying speech intelligibility deficits associated with dysarthria.

Idioma originalEnglish
Título de la publicación alojadaSpeech and Computer - 23rd International Conference, SPECOM 2021, Proceedings
EditoresAlexey Karpov, Rodmonga Potapova
Páginas670-679
Número de páginas10
DOI
EstadoPublished - 2021
Evento23rd International Conference on Speech and Computer, SPECOM 2021 - Virtual, Online
Duración: sept 27 2021 → sept 30 2021

Serie de la publicación

NombreLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volumen12997 LNAI
ISSN (versión impresa)0302-9743
ISSN (versión digital)1611-3349

Conference

Conference23rd International Conference on Speech and Computer, SPECOM 2021
CiudadVirtual, Online
Período9/27/21 → 9/30/21

Nota bibliográfica

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Financiación

Acknowledgments. This work was supported by National Institutes of Health under NIDCD R15 DC017296-01.

FinanciadoresNúmero del financiador
National Institutes of Health (NIH)NIDCD R15 DC017296-01

    ASJC Scopus subject areas

    • Theoretical Computer Science
    • General Computer Science

    Huella

    Profundice en los temas de investigación de 'Increasing the Precision of Dysarthric Speech Intelligibility and Severity Level Estimate'. En conjunto forman una huella única.

    Citar esto