Analysis And Evaluation Of Functional Articulation Disorder Based On Speech Recognition

Preschool period is the key period for children's speech and language development.In China,the prevalence of children's articulation disorder is about 2.14%.However,at present,the effective assessment methods mainly focus on the subjective method of auditory perception.The objective analysis methods lack attention,and the analysis and evaluation results lack objectivity,accuracy and stability.These problems result in many preschool children losing the best opportunities for correction and rehabilitation.With the development of speech recognition technology,the application of speech recognition technology has covered many fields,especially in speech education and daily consumer life.However,in the analysis and evaluation of children's articulation disorders,the relevant research results and applications based on speech recognition at home and abroad are few,and researchers have not paid enough attention to it.According to the research status and development trend of speech articulation and speech disorder assessment methods at home and abroad,based on speech recognition technology,this paper realizes the analysis and evaluation of children's functional articulation disorder.The research contents and results are as follows:According to the cognitive ability of 3-6 years old children,a set of assessment tables for dysarthria was designed.200 healthy children and 67 children with dysarthria were asked to follow the content of the assessment form and collect voice at the same time.Using HTK acoustic modeling tool based on hidden Markov model and SRIM language modeling tool to model children's data respectively,the recognition rate of the final recognition model is 73.12%.Using the established recognition and evaluation model,the preliminary analysis and evaluation of children's functional dysarthria were successfully realized.In order to prevent the low speech recognition rate caused by the lack of training sample data and the inaccurate evaluation results,the GOP(The Goodness of Pronunciation)algorithm is added to improve the evaluation model.GOP algorithm evaluates the probability of phonemes corresponding to the target speech segment.The higher the coincidence between the target speech and the corresponding phonemes,the more accurate the phoneme pronunciation is.On the contrary,the lower the probability,the greater the deviation between the target speech and phoneme,and the worse the pronunciation.The improved system is used to carry out comprehensive analysis and evaluation of children's functional articulation disorder in four aspects: clarity,fluency,accuracy and articulation ability.It can get rid of the way that only one item of articulation ability was used toevaluate children's functional articulation disorder to reflect the severity of children's functional articulation disorder,so as to make the evaluation results more comprehensive and accurate.