# Systematic Review, Meta-Analysis And Meta-Synthesis Of Automatic Classification Of Stuttering Through Machine Learning Authors: L.Barrett, J. Hu, & P.Howell ## Background Modern speech recognition systems are increasingly accurate and useable with fluent speech. However, systems often fail to capture dysfluent speech appropriatly, resulting in a loss of accuracy and useability. More research groups are trying to address this issue by designing classification models to separate stuttered from fluent speech. These models can be used both in the clinical assessment of stuttering but also to improve current automatic speech recognition system. Thus far, to our knowledge, there are no systematic reviews or meta-analysis investigating the state of the field. With more interest in the field, it is important to review the current advantages and possible limitations of automatic classification of stuttered speech. This project aims to review the extant literature, estimating the degree to which models are able to classify stuttering. This will be assessed both through systematic review of the literature but also, through a meta-analysis and -synthesis of the data. ## Review Methods ### Review question(s) 1. How does model performance vary as a function of model type across the field? 2. How does model performance vary as a function of the input features to the model? 3. How does model performance vary as a function of database size? 4. Does database size, input features and model types interact in their resultant model performance? ### Search protocol This can be found under the files tab of the project. ### Context #### The review: - All forms of machine learning are considered. - Only data from people who stutter will be considered. Data from people with neurogenic stuttering is to be excluded. No constraints are set on the features extracted from the speech signal. Additional meta-data such as, stuttering severity and subject identifier will also be included. ## Additional Information A full description of the search terms, data extracion methods and meta-analysis is beyond the scope of the Wiki. A full document is provided in the files.