Speech Recognition for Hindi using Zero Crossing Rate Methods

Speech Recognition is the most promising field of research and technology. Speech to text conversion is the process of converting input acoustic speech signal into the text similar to information being conveyed by the speaker. This paper is to build speech to text conversion system for Hindi language to reduce the gap between computer and people in rural areas. Although there are many interfaces are already available, but need is to build more vocabulary and accuracy in it. The system is trained for 100 words, collected from different speakers of different age groups. In this system features including zero crossing rate and short term energy is studied. In this paper mainly three phases are there, training phase, testing phase and recognition phase. In training phase, training database is created with Hindi speech samples and trained using feature extracted using algorithms of zero crossing rate and energy calculation. In testing phase extracted features are matched and then created testing database with minimum and maximum range of zero crossing rate and energy from the speech samples. In recognition phase same techniques applied to speech and compared with the database and fetch word using data of training and testing phase.