Feature Extraction For Speaker Recognition
Identification of the people in today’s world is important than ever. We have many bio-metric methods to so this
like fingerprint, face recognition, retina scan etc. these techniques are capable of identifying only one person at a time. If
bio-metric information of two or more people is mixed together then these methods fail catastrophically. Also in these methods
require the user to be in a specific position for a specified time, which can get very tiresome and hectic if large number of
people are to be identified. In this paper we focus on the identification of people based on speech. Also we deal with the
problem of multiple people taking a single instant. We base our research of project on text-dependent model and then we even
try to analysis the text-independent identification. First, speaker signal will go to pre-treatment process, where it will remove
the background noise if any and required. Then, features from speech signal will be extracted using Cepstrum domain. We get
very promising results even at moderate noise levels. The resemblance of features used for database and for testing is good.
Index Terms—Bio-metric, Cepstrum, text-dependent model, text-independent model.