HMM-Based Speaker Gender Recognition for Bodo Language

Speech, the act of speaking, is the most natural way of exchanging information between homo sapiens. Speech primarily conveys the message via words, spoken by the speaker. Speech also conveys the emotion with which the speaker speaks, speaker’s health condition, gender of the speaker, and also the language in which the speaker is speaking. Systems which aim to recognize the speaker-related information in speech signals through an extraction and characterization process are called speaker recognition systems. Speaker recognition applications are becoming common and useful nowadays as many of the modern devices are designed and produced for the convenience of the general public. Speaker recognition systems are developed for many indigenous languages. Application of hidden Markov models (HMMs) to speaker recognition has seen considerable success and gained much popularity. This paper presents an attempt made toward developing a speaker gender recognition system. A model built using Hidden Markov Model Toolkit (HTK 3.4.1) has been trained and tested on sample speech of either gender in Bodo language, and results show good recognition.