Automatic Speech Recognition (ASR), in short Speech recognition or speech-to-text is the process of converting an acoustic signal, captured by a microphone or a telephone, to a set of words. Language model is one of the curial datasets required to conduct Automatic Speech Recognition. Most of the language models in use for speech recognition are developed based on standard n-gram modeling. In this research, we have attempted to develop a maximum entropy language model for speech recognition. A two pass attempts are made in this research, one pass using word-form based language model and the other pass using the morpheme-based language model for speech recognition. Our model performs better as the n-gram constraint increases from bigram to five-gram both in wordform and morpheme-based language models. Also at the increased n-gram constraints, our model performs better than standard n-gram model. It is observed that, the morpheme based would be a much more promising model if the morphological segmentation is manually supported.