Text to Speech Conversion of Handwritten Kannada Words Using Various Machine Learning Models

Abstract

Recognition of handwritten characters and words is challenging due to the presence of complex character sets and the complexity of the words. The machine learning models with feature extraction methods will help us to solve the problem of recognizing handwritten words. The various preprocessing techniques applied to the word are Bilateral filters, resizing the images to find the Region of Interest (ROI) by contour detection and cropping the images. After resizing the image, it is further deskewed for better results. The recognition of handwritten Kannada words by extracting histogram of oriented gradients (HOG) features from the word image using various Machine Learning (ML) techniques are presented in this paper. Then the recognized word is converted to speech using the Google Text-to-Speech (gTTS) API. The dataset consists of 54,742 handwritten word images. Various machine learning models like Support Vector Machine (SVM), k-nearest neighbors (KNN), and random forest were applied to the dataset. Average accuracy of 88% is obtained using the SVM classifier with Radial Basis Function (RBF) kernel.