Isolated Telugu Speech Recognition on FWT and HMM based DNN Techniques

Abstract

Automation is dramatically changed in the present technology. Even in the small villages they are using advanced technology. This paper deals with automatic speech recognition where a local language Telugu can be recognized by the system, the human machine interaction is easy if this recognition is perfect. There are many advanced techniques to design such systems but every time the procedure is different to obtain the promising results. This research uses suitable techniques like FWT for features extraction and HMM based DNN for feature classifications. The speech copra is trained and tested on various types of speech frequencies which deal with different parameters.  The research used isolated words for recognition, where most frequently used 50 words are recognized. This is performed for speaker independent model.