While most people in the world speak more than one language, it has been reported that the performance of existing monolingual speaker identification deteriorates when there is a language mismatch. In this study, we examine whether speakers are correctly identified when the languages of the training set and test set are changed for 13 Japanese-English bilinguals of both genders. The experiment was conducted using an x-vector calculated from 20-dimensional log MFBs as features. The results showed that the accuracy was higher when the training set was English and test set was Japanese than the opposite condition, regardless of the lengths of the input segment, although it was inferior to monolingual speaker identification. When we conducted similar experiments by removing some of the dimensionality of the log MFBs, we found that there were differences in accuracy depending on the language in which they were trained, indicating that the frequency bands with the most important amount of information differ among the languages.