Machine Learning Model Based Speaker Recognition

Authors

  • Dr. G. Chenchamma, Nagendra babu Rajaboina,K. Prasuna

Abstract

Machine learning is the method of data analysis which constructs analytical models automatically. Machine
learning uses iterative algorithms to learn from data and allows the computer to find information, hidden
values that are not explicitly programmed. The repetitive aspect of Machine learning is important because
when these models are presented to new information, they can adjust freely. Speech is one of the most basic
forms of communication among humans. Speech is a signal that carries information about the message to be
conveyed, the characteristics of the speaker and the language of communication. Every person has a unique
voice by which that person can be recognized. Programmed Speaker Recognition (ASR) is assignment of
perceiving an individual from his/her voice by a machine. This Paper defines the Machine learning based
Speaker recognition or a pattern recognition task which involves three phases namely, feature extraction,
training and testing. In the feature extraction stage, features representing speaker information separated from
the discourse signal. By utilizing the LPCC, MFCC, LP lingering and mix of source (LP Residual) and
framework (LPC's or MFCC's) got from the discourse information is utilized for preparing and testing. In
the training phase Gaussian mixture models are built, one for each speaker, using the training data of the
speaker. Throughout the testing phase, the models are tested with the test data. Based on the results with test
data, decision is made about the identity of the speaker Speech Recognition.

Published

2020-02-28

Issue

Section

Articles