A Novel Stuttering Disfluency Classification System Based on Respiratory Biosignals

Stuttering is the principal fluency disorder that affects 1% of the world population. Growing with this disorder can impact the quality of life of the adults who stutter (AWS). To manage this condition, it is necessary to measure and assess the stuttering severity before, during and after any therapeutic process. The respiratory biosignal activity could be an option for automatic stuttering assessment, however, there is not enough evidence of its use for this purposes. Thus, the aim of this research is to develop a stuttering disfluency classification system based on respiratory biosignals. Sixty-eight participants (training: AWS=27, AWNS=33; test: AWS=9) were asked to perform a reading task while their respiratory patterns and pulse were recorded through a standardized system. Segmentation, feature extraction and Multilayer Perceptron Neural Network (MLP) was implemented to differentiate block and non-block states based on the respiratory biosignal activity. 82.6% of classification accuracy was obtained after training and testing the neural network. This work presents an accurate system to classify block and non-block states of speech from AWS during reading tasks. It is a promising system for future applications such as screening of stuttering, monitoring and biofeedback interventions.