Research Results of Artificial Neural Network for User Authentication According to Frequency of Forth Formant of Vowel Sound Phoneme

In this article it is suggested using the frequencies of the forth formant of the vowel phoneme as speech parameters for voice authentication. The differences of the meanings of the words “sound” and “phoneme” are given. The phoneme peculiarities of vowel and consonant sounds are examined. The results of 2 experiments confirming the expedience of using these parameters are indicated. During the experiments each of the six users pronounces sounds of all vowel phonemes 10 times. The artificial neural network is created for user authentication as per these speech parameters. 36 training sets are developed for it, 6 sets for each of the users. The efficiency control of the artificial neural network operation is performed according to the data of the first two experiments. 1st type error (FRR - False Rejection Rate) and 2nd type error (False Acceptance Rate), which are further compared with the corresponding errors obtained as a result of the authentication according to the other speech parameters, are revealed.