Voice Recognition and Gender Classification in the Context of Native Languages and Lingua Franca

Voice verification and gender classification from voice were carried out in the context of native (mother tongue) languages and lingua franca languages. A total of 3980 voice utterances recorded in English language and 28 native languages were acquired from 520 bilingual subjects in this paper. We first determined the cross linguistic influence of mother tongue by bilingual speakers on the verification performance of voice recognition using English and native languages' gallery and probe sets. Secondly, we employed transfer learning in training four convolutional neural network models for classifying gender from voice, using training and test samples of English language, exclusively; one dominant native language; and a mixture of 28 native languages. Our results do show that mother tongue or first language, intonation variations, language variety in the training or test sets do influence voice verification and gender classification.