Emotional States Discovery under Multiple Disguising Effects for Speaker Identification

Today“s technology is rapidly evolving towards the discovery of abnormal voices since voices are not neutrally spoken most of the time. Introducing disguising effects to abnormal voices has become a crucial threat in various activities. As a result, this would have a considerable impact on the Speaker Identification (SI) performance. Therefore, identifying the unknown speaker in emotional talking conditions under multiple disguising effects is a fundamental concern to be exploited in this research. Six distinct emotional talking conditions along with three effects, i.e.., low-pitched, high-pitched, and Electronic Voice Conversion (EVC) are used to evaluate the performance of speaker identification. Furthermore, this study employs two different speech databases, which are a global English dataset called Speech Under Simulated and Actual Stress (SUSAS) and an Arabic Emirati-accented database. Results reveal that the average speaker identification performance in neutral talking condition is the highest under low-pitched effect with a 96% accuracy. On the other hand, the average reported performance for all of the emotional states under the same effect reaches 81.7%.