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 original | English |
|---|---|
| Título de la publicación alojada | Speech and Computer - 23rd International Conference, SPECOM 2021, Proceedings |
| Editores | Alexey Karpov, Rodmonga Potapova |
| Páginas | 670-679 |
| Número de páginas | 10 |
| DOI | |
| Estado | Published - 2021 |
| Evento | 23rd International Conference on Speech and Computer, SPECOM 2021 - Virtual, Online Duración: sept 27 2021 → sept 30 2021 |
Serie de la publicación
| Nombre | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volumen | 12997 LNAI |
| ISSN (versión impresa) | 0302-9743 |
| ISSN (versión digital) | 1611-3349 |
Conference
| Conference | 23rd International Conference on Speech and Computer, SPECOM 2021 |
|---|---|
| Ciudad | Virtual, Online |
| Período | 9/27/21 → 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.
| Financiadores | Número 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
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver