Sistem Pengenalan Dan Verifikasi Pembicara HMM

Budi, Darmawan and Suthami, Ariessaputra (2018) Sistem Pengenalan Dan Verifikasi Pembicara HMM. Prosiding CITEE 2018. pp. 68-73. ISSN 2085-6350

Camera Ready.pdf

Download (568kB) | Preview


the weakness of the HMM speaker recognition system is to group all inputs into one of the speakers contained in the database, although the voice input does not include the voice of one of the speakers listed in the system database, and this is a mistake. In this study, the HMM speaker recognition system will be integrated with HMM speaker verification system so that a person's non-voice voice signal from one of the speakers contained in the database will be rejected by the system. In this study will also be examined some number of states and some threshold value used to see the number of state and threshold value which is best for use on the system. Voice signal data is the result of voice recording obtained from twenty speakers. Ten speakers will be built HMM model and will be entered into the system database. Each of these ten pronouns the word "HADIR” 30 times. 10 pieces of voice cues will be used in the training process and 20 other voice cues are used for the testing process. As for the other 10 speakers say the same word 20 times that will be used in the testing process. The results of the experiment show that the best recognition system is in the HMM model with the number of state 4, the value of% threshold -4% and with an average accuracy of 73.75%.

Item Type: Article
Keywords (Kata Kunci): Hidden Markov Models; Speaker recognition; verification, MFCC.
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions: Fakultas Teknik
Depositing User: Budi Budi Darmawan
Date Deposited: 14 Nov 2018 02:24
Last Modified: 14 Nov 2018 02:24

Actions (login required)

View Item View Item