Lisp detection and correction based on feature extraction and random forest classifier

Abstract

Lisp is a functional speech impediment that results in difficulty to produce specific speech sounds and specific words. The objective of this paper is to delineate a compound speech processing algorithm that can segment and recognize the individual words present in the speech using feature extraction, and identify any lisp words using Random Forest Classifier and correct it. The features extracted are the Mel Frequency Cepstral Coefficients (MFCC). The coefficients are extracted and they form the basis of classification into lisp or non-lisp words. MRF Algorithm has been proposed (MFCC-RF) that can be applied on real-time embedded systems that can help people with speech disability. The proposed model can be used in various speech to text applications as the algorithm detects lisp words accurately and correct them in real time. Different classification algorithms such as the regression algorithm and Fuzzy Decision Tree classification algorithm have been used and their results have been compared. It has been observed that the Random Forest Classifier gives superior performance.