Can a syntactic language model be incorporated into an HMM-based ASR framework?

Automatic speech recognition (ASR) is the process which a computer interprets human speech into some kind of meaningful representation. Usually, the task for the automatic speech recognition is to identify the correct wordsequence. Speech recognizer is used in different applications for example simple command-control programs, transcription and speech understanding. A good automatic speech recognizer should be able to recognize spontaneous continuous speech and should not require the speaker to break up their speech into discrete words. The task of recognizing speech from newspapers or news broadcast with a state of the art speech recognition systems can be done with accuracy higher than 90 percent. The accuracy drops significant when using spontaneous speech, due to the fact that the acoustic and language models usually have been built using written language or speech from written language [9]. This term paper is a literature review in the area of automatic speech recognizer and we will investigate different techniques to incorporate syntactic information into the speech recognizer. How can a speech recognizer benefit from syntactic information? Is it possible to integrate a parser? The paper is organized as follows. We begin with a brief overview of automatic speech recognition. Section 3 investigates different methods of integrating language models. Finally, the paper ends with conclusions.