The Promise of Automatic Speech Recognition for Improving Early Identification of Spanish-English Bilingual Children At-Risk for Language Disorders
Abstract/Description/Artist Statement
Children with language disorders often struggle academically and socially, making rapid identification and intervention critical to support long-term outcomes. Bilingual children, in particular, are at an increased risk for being misidentified with language disorders: a direct result of inadequate staffing of bilingual clinicians and limitations in valid and normed assessment tools. Fortunately, emerging advancements in automatic speech recognition (ASR) software have the potential to ameliorate these challenges by using artificial intelligence to transcribe language samples. The purpose of this multi-university study was to evaluate the transcription accuracy of ASR software for Spanish-English bilingual children.
A team of ODU researchers scored 190 audio recordings of 5-7-year-old Spanish-English bilingual children by using the Spanish Screener for Language Impairment in Children (SSLIC; Restrepo et al., 2013). The SSLIC audio recordings covered five specific elements of speech: articles, prepositions, derivation, clitics, and subjunctive tense. Manual scoring provided a baseline for data comparison with the scores generated using the ASR software. Following the ODU scoring, the audio recordings were transcribed using ASR software, then analyzed using an automated scoring algorithm that assessed the presence of the previously established elements of speech.
Preliminary results suggest no significant difference between SSLIC's manual and ASR scoring. This demonstrates the potential of artificial intelligence tools in the field of communication sciences and disorders, opening up an opportunity to enhance assessment and therapy tools for bilingual children with language disorders. Further research is necessary to investigate the benefits and limitations of this emerging technology.
Faculty Advisor/Mentor
R.J. Risueño, PhD CCC-SLP
Faculty Advisor/Mentor Email
rrisueno@odu.edu
Faculty Advisor/Mentor Department
Speech-Language Pathology
College/School Affiliation
Ellmer College of Health Sciences
Student Level Group
Undergraduate
Presentation Type
Poster
The Promise of Automatic Speech Recognition for Improving Early Identification of Spanish-English Bilingual Children At-Risk for Language Disorders
Children with language disorders often struggle academically and socially, making rapid identification and intervention critical to support long-term outcomes. Bilingual children, in particular, are at an increased risk for being misidentified with language disorders: a direct result of inadequate staffing of bilingual clinicians and limitations in valid and normed assessment tools. Fortunately, emerging advancements in automatic speech recognition (ASR) software have the potential to ameliorate these challenges by using artificial intelligence to transcribe language samples. The purpose of this multi-university study was to evaluate the transcription accuracy of ASR software for Spanish-English bilingual children.
A team of ODU researchers scored 190 audio recordings of 5-7-year-old Spanish-English bilingual children by using the Spanish Screener for Language Impairment in Children (SSLIC; Restrepo et al., 2013). The SSLIC audio recordings covered five specific elements of speech: articles, prepositions, derivation, clitics, and subjunctive tense. Manual scoring provided a baseline for data comparison with the scores generated using the ASR software. Following the ODU scoring, the audio recordings were transcribed using ASR software, then analyzed using an automated scoring algorithm that assessed the presence of the previously established elements of speech.
Preliminary results suggest no significant difference between SSLIC's manual and ASR scoring. This demonstrates the potential of artificial intelligence tools in the field of communication sciences and disorders, opening up an opportunity to enhance assessment and therapy tools for bilingual children with language disorders. Further research is necessary to investigate the benefits and limitations of this emerging technology.