Classification of Stuttered Disordered Speech and Normal Speech of Marathi Language by using ANN

Abstract

Stuttering is one of the WHO's recognised complex disorders. Speech disorder analysis has recently become one of the growing are of research in the field of Digital Speech Signal Processing and natural language processing.  We have created our own stuttered speech database for Marathi language. This is the first database developed for Marathi language with 8,481 utterances. This paper describe automatic classification of stuttered speech and normal speech. Mel Frequency Cepstral Coefficients (MFCC) feature extraction is implemented for framing, windowing and extracting feature by using Librosa python library and MLP is used for classification purpose. Result shown that individual accuracy for stuttered (99.54%) and normal speech (95.83%) with overall average accuracy 97.68%. This work is developed by using open source python libraries.