A Game with a Purpose for Automatic Detection of Children’s Speech Disabilities using Limited Speech Resources

Reem Salem, Mohamed Elmahdy, Slim Abdennadher, Injy Hamed


Abstract
Speech therapists and researchers are becoming more concerned with the use of computer-based systems in the therapy of speech disorders. In this paper, we propose a computer-based game with a purpose (GWAP) for speech therapy of Egyptian speaking children suffering from Dyslalia. Our aim is to detect if a certain phoneme is pronounced correctly. An Egyptian Arabic speech corpus has been collected. A baseline acoustic model was trained using the Egyptian corpus. In order to benefit from existing large amounts of Modern Standard Arabic (MSA) resources, MSA acoustic models were adapted with the collected Egyptian corpus. An independent testing set that covers common speech disorders has been collected for Egyptian speakers. Results show that adapted acoustic models give better recognition accuracy which could be relied on in the game and that children show more interest in playing the game than in visiting the therapist. A noticeable progress in children Dyslalia appeared with the proposed system.
Anthology ID:
W17-7704
Volume:
Proceedings of the 1st Workshop on Natural Language Processing and Information Retrieval associated with RANLP 2017
Month:
September
Year:
2017
Address:
Varna, Bulgaria
Editors:
Mireille Makary, Michael Oakes
Venue:
RANLP
SIG:
Publisher:
INCOMA Inc.
Note:
Pages:
28–34
Language:
URL:
https://doi.org/10.26615/978-954-452-038-0_004
DOI:
10.26615/978-954-452-038-0_004
Bibkey:
Cite (ACL):
Reem Salem, Mohamed Elmahdy, Slim Abdennadher, and Injy Hamed. 2017. A Game with a Purpose for Automatic Detection of Children’s Speech Disabilities using Limited Speech Resources. In Proceedings of the 1st Workshop on Natural Language Processing and Information Retrieval associated with RANLP 2017, pages 28–34, Varna, Bulgaria. INCOMA Inc..
Cite (Informal):
A Game with a Purpose for Automatic Detection of Children’s Speech Disabilities using Limited Speech Resources (Salem et al., RANLP 2017)
Copy Citation:
PDF:
https://doi.org/10.26615/978-954-452-038-0_004