A Review of Robust Deep Learning-Based Speaker Recognition

Speaker identification is the process of identifying individuals through their vocal characteristics. Over the past few years, the efficiency of speaker identification systems has undergone a remarkable transformation, largely owed to the advancements in deep learning methodologies. Nevertheless, as speaker identification technology continues to advance, the focus has shifted towards recognizing speakers in real-world situations. These scenarios often involve numerous domain disparities and noise challenges that need to be addressed. In the past few years, deep learning has excelled in domain adaptation and noise reduction, further optimizing speaker recognition methods. Utilizing the robust speaker recognition approach of deep learning, the ongoing investigation is categorized into three distinct areas: domain adaptation, speech enhancement, and data augmentation. This classification is determined by the role of deep learning in speaker recognition, aiming to analyze and summarize the characteristic algorithms and network structures employed in each area. Furthermore, a comprehensive comparison and analysis of their respective performance is conducted. In conclusion, this section examines the challenges and issues encountered in the present speaker recognition research.