The attention of researchers is directed towards the generation of expressive speech. Hence the text which would enter the text-to-speech system must be such that, the sentiment which it encompasses is to be known. So that this sentiment analyzed text can be then used to synthesize expressive speech. This paper concentrates on identifying and categorizing the sentiment present in text, which is the input to the text-to-speech system, which is the preliminary step for the production of expressive speech. The sentence is preprocessed and classification is done using the classifiers and sentiment tagged sentence is passed as input to the text-to-speech system and text-to-speech system selects the voice based on the sentiment and converts text to expressive speech. The twitter corpus is used as the dataset for training and testing the classifier because of its affect based expressiveness.