Speaker Identification in Different Emotional States

A major challenge degrading the robustness of speaker-recognition systems is the variation in the emotional state of speakers. In this study, we propose a speaker recognition system in an emotional state for two languages, Arabic and English. In addition, cross-language speaker recognition was applied. Convolutional neural network (CNN) and long short-term memory (LSTM) models were used to design a convolutional recurrent neural network (CRNN) main system. The overall CRNN system exhibited an accuracy as high as 97.5% and 97.18% for Arabic and English emotional speech inputs. For the cross-language program, the overall accuracy was as high as 91.83%.