Complex Channel Speaker Recognition Technology

After decades of development, computer science has produced a significant impact in human's life and work. With personal computers, smart appliances, mobile operating system technology widely used, people put forward higher requirement in HCI(human-computer interaction).speech technology is considered as an ideal HCI, and is widely used in telecommunications, education, security and other industries.Speaker recognition is an important branch of speech processing; its basic principle is to "know" the people through analysis and modeling in speaker's voice characteristics. Because the Gaussian Mixed Model can well fit in characteristics of speaker, with lower Complexity, it has been a mainstream solution to speaker recognition. Based on GMM, Reynolds and others developed Gaussian Mixed-Universal Background Model (GMM_UBM) to solve insufficient training problem in GMM.The main research of this article is robust speaker recognition in complex channel, aim to enhance the performance in non-ideal environment. Paper mainly focuses on three main aspects of recognition system--feature extraction, speaker modeling, similarity computing, the popular channel-robust method and algorithms is discussed to enhance the ability to adapt to different environments.Robot has bright future, family-oriented service robots is strong demand for millions of families. At the last chapter of paper, discuss a speaker recognition sub-system embedded in a family-oriented service robot named "RuiBao", in the process between human-robot interaction, the sub-system identify and verify the speaker's authority intelligently to ensure the security.