| Abstract |
Speech and language difficulties in children, how they are assessed and identified within the education system, and the interventions available to them, have been a topic of hot-debate in recent years (see the Bercow Report, 2008, 2018; Department for Education, 2012; The Marmot Review, 2010). Not identifying children early on can have detrimental life-long consequences, for example, those with speech and language difficulties are over 5 times less likely to attain 5 A*-C grades in their GCSEs, account for up 60% of young offenders, and 88% of unemployed men (Department for Education, 2012). Yet, developments in research and impact on Policy has plateaued in the last ten years ("The Bercow Report: Ten Years On", Bercow, 2018). The proposed research aims to develop and validate an easy-to-use and easily accessible App-based speech assessment, to identify children aged 4-6 years, who may be at risk of speech difficulties. To achieve this goal, the project aims to train an automatic speech recognition (ASR) algorithm to separate spoken words into syllabic units, identify syllable-based phonological errors and score disfluencies in speech, thereby automating the analysis of speech samples. A large sample (~1000) of speech is needed in order to achieve this, and the sensitivity and specificity of the automated assessment will be verified against a clinical gold standard. Incorporation of this automated speech assessment into an existing web-application, assessing language, would provide educational practitioners with a valuable tool for identifying children at-risk of speech, language and communication disorders at a young, and potentially critical, age. The research findings could have widespread and significant implications for a variety of stakeholders and beneficiaries. Once validated, the assessment could be used on a large scale as a cheap, rapid and reliable tool for screening children upon school entry, which, considering the devolution of budgets from local authorities, would be highly beneficial to schools and academies. Referrals made by schools based on an objective and scientifically-validated measure would also benefit and increase the efficiency of speech and language intervention. The ASR algorithm would be an advantageous tool for many academic and non-academic researchers interested in speech analysis. An up-to-date estimate of speech difficulty prevalence would be useful for the government, policy-makers and researchers alike. The benefits surrounding the increased reliability in identification of at-risk children, could allow for earlier intervention and thus, improved outcome post-intervention, which will improve their later educational attainment, opportunities, and overall quality of life. Completion of this project would also benefit academics, both in English- and non-English speaking countries. The ability to train the ASR algorithm with a variety of languages increases the potential impact of the research outcomes and provides the opportunity to expand its use to other countries and contexts. Additionally, the findings can be used as a basis for investigations by other researchers, helping to progress research in the interdisciplinary fields of speech and language sciences, clinical assessment, computational modelling and machine learning. Automated transcription of syllable boundaries would reduce the need for manual transcription, saving resources and speeding up data analysis for researchers in the future. |
