Large amount of information are embedded in natural scenes. Signs are good example of natural objects with high information content. Visually impaired individuals are unable to utilize the significant amount of information in signs. This paper presents a system for detecting and recognizing the signs in the environment specially written in Bengali and voice synthesizing their contents. In this paper we present an algorithm for detection and localization of text in road sign using edge detection approach. We use an adaptive thresholding method to binarize the text blocks and recognize the text using a neural network based OCR module. Finally, the recognized text is synthesized as voice output message convenient for visually impaired. The detection, recognition and speech synthesis modules each perform well on their respective task, and initial evaluation of the complete system is promising.