Mem2Speech: An Intelligent Character Recognition-to-Speech Application Using Long Short-Term Memory Networks

Abstract

Smartphones are no longer used solely for communication but also for photography, video recording, internet sur fing, etc. As technology continues to advance, it is possible to apply some techniques to perform tasks such as Optical Character Recognition (OCR) and Intelligent Character Recognition (ICR). Tess2Speech is one of the applications that perform these tasks and it uses the Tesseract OCR Engine. Each trained Tesseract model of Tess2Speech is able to recognize a handwriting style of a user but not that of other users. It is possible for each user to train the engine themselves to t their handwriting but this may be an issue if they don't have the technical background needed to train the engine. This special problem uses Long Short-Term Memory (LSTM) networks for handwritten text recognition. It provides a user-friendly trainer that will help AI experts in training handwriting recognition models. It also provides a mobile application for recognizing handwritten texts using the trained models.