Speech recognition using HTK toolkit for Marathi language

Marathi language is one of the oldest languages in India, mostly used in Maharashtra region. In this paper, we have implemented Automatic Speech Recognition system for Marathi language using HTK Toolkit. In the ASR process, computer takes a voice signal as an input, which is recorded using a microphone and converts it into sequence of words. Our ASR is a machine, with an ability to recognize a recorded Marathi speech signal (voice commands). The HTK i.e. Hidden Markov Model Toolkit is used to develop the proposed ASR System, which we have trained using GMM-HMM based Acoustic models for mono-phones and created bi-gram language model. At the time recognition these trained models are used to recognize the uttered words. In this work, training data contains 910 Marathi sentences, recorded by 3 male and 3 female speakers.