Persian Language telephone and Microphone Speaker identification using neural networks

Abstract

Speaker identification has attracted much attention as an identity identification method in security systems. Identity identification based on telephone conversations may be very useful in guaranteeing personal information security. Nowadays, virtual neural network algorithms are highly used in machine learning. These networks are trained by labeled data and they perform classification. In this study, speaker identification systems based on Mel Frequency Cepstral Coefficients are designed for Persian language speakers. The results of two tests have been reported: Cepstral Coefficients and neural networks are used in the first test to detect microphone speaker; the data base used in this test is the authenticated data of Farsdat, and we used 7 speakers in them. The second test performs the telephone speaker identification for 10 Persian language speakers using lDC data; this test used the Persian language speakers’ part of this database. The best results in microphone mode are obtained having 15 minutes training data and the proposed system achieved the identification rate of 100 percent. The best identification result with 3 minutes of training data achieved 82 percent