Speech Recognition System for Malayalam Language

Abstract

Speech is a dynamic process without clearly distinguished parts. The development of a speaker independent, continuous, automatic speech recognition system is crucial for any language, as it could make human-machine communication easier. Few works can be found in the Malayalam speech recognition field but a fully functional, continuous, automatic recognition system is yet to be developed. The speech engine used here is the CMU-Sphinx system, an open source continuous speech, speaker independent recognition system developed at Carnegie Mellon University. The sphinx decoder requires an Acoustic model and a Language Model. An acoustic model is the statistical representation of the speech units. Language model is the probability distribution over sequences of words. Pocket sphinx libraries are used here, which helps in further extending the recognition to any portable platforms such as android.