Analysis of Acoustic Features in Gender Identification Model for English and Bahasa Indonesia Telephone Speeches

摘要

One of the most interesting topics in auditory problem is determining gender of the speaker. In recent years, machine learning has gained significant attentions as a way to build a classifier from labeled data which also can be implemented to build a gender classifier. In this study we develop gender classifier using two different datasets with different languages, English and Bahasa Indonesia. Each data from both datasets is represented by 20 acoustic features. Multi Layer Perceptron (MLP) is used to build the classification model using all these features and trained only on English dataset. This model is evaluated in both dataset to get the performance matrices consist of accuracy, AUROC, precision and recall. Ultimately, using this model we also identify and compare important features from both dataset to see the different characteristics of English and Bahasa Indonesia speeches.