In speaker recognition, a way to identify individual's identity is to identify the characteristics of the individual voice, which can be identified using speaker's speech embedding. In this paper, we propose a method using x-vector and d-vector speech embedding feature extraction and ResNet50 model on identifying user's identity. This paper uses part of VoxCeleb2 dataset with total train data of 12.207 utterances, total validation data of 3.618 utterances, and total testing data of 2.068 utterances from 552 speakers. Research done in this paper shows ResNet50 model with x-vector achieved accuracy, recall, precision, and f-1 score of 74.12%, 73.42%, 78.02%, and 0.72 respectively, whereas ResNet50 model with d-vector achieved accuracy, recall, precision, and f-1 score of 36.17%, 35.77%, 36.86%, and 0.32 respectively, whereas the current state-of-theart model achieved accuracy, recall, precision, and f-1 score of 10.97%, 10.61%, 6.64%, and 0.07 respectively.