Abstract
SPEAKER AUTHENTICATION USING ZERO CROSSING RATE WITH RESPECT TO BODO VOWEL PHONEME: A CLASSICAL EXPERIMENT
Bimal Kumar Kalita* and Dr. (Professor) Pran Hari Talukdar
ABSTRACT
The researcher fraternity in the field of Natural Language Processing have been always enthusiastic about the possibility of machine human interaction. In this area authentication of a certain person based on speaker?s specificity in his/her voice is always attracting a section of scientists. Speaker authentication is a process of recognition and verification of the identity of a claimant speaker through the properties of his/her utterances i.e. analyzing the voice characteristics and then come to a conclusion whether the claim of the identity is correct or he/she is an imposter matching the pre existing data in the database. The Bodo language belongs to the Tibeto-Burman language family, which is a sub-group of the Sino-Tibetan language group. It is one of the popular Indian tribal languages, primarily spoken in Assam, the north-eastern state of India. Speakers are also found in adjoining areas of Assam and border areas of West Bengal and Bangladesh. In our paper experiments are performed with the method of speaker dependent modelling. System used consists of three phases. i. Training ii. Testing and iii. Recognition/Authentication. Five female and male speakers each are taken for speech database. Phoneme utterances are recorded with ten repetitions for feature extraction. Zero Crossing Rate is a feature vector which can be useful as an acoustic metric. In our study ZCR feature extraction of Bodo vowel phonemes are performed frame by frame. ZCR analysis will be helpful in designing a BODO Speaker Recognition, Identification and Authentication System (SPERIA-B i.e. SPEaker Recognition, Identification and Authentication for Bodo).
[Full Text Article] [Download Certificate]