Hindi speech synthesis by concatenation of recognized hand written devnagri script using support vector machines classifier

Handwritten optical character recognition is one of the major research area due to its complexity in segmenting the character which increases in the case of Devnagri Script due to Modifiers and compound characters. Thus this paper shows an adaptive segmentation technique which shows less error. This paper shows an implementation of Handwritten Devanagri character recognition system. Keeping in mind the impairment of blind people, OCR system is extended to Text to Speech System. As by the surveys Support Vector Machines found to be very efficient and robust in handling large amount of features, SVM was used for the purpose of classification. The learning was done in Multi domain feature. In Transform domain, wavelet transform was used because of its ability to keep global feature in different scale. In structural domain, gradient and distance profile features were used as they represent local characteristics of the characters. Here the handwritten document was segmented adaptively in 3 levels; line, word and character. Preprocessing was done using various Morphological operation and thinning techniques. All the training was done using self-created database consists of 50 samples per character from 10 different individuals. Further, the recognized characters were digitized using Unicode and their corresponding phoneme present in the database were concatenated to form the speech signal. A self-created database of all the possible phonemes was created.