Dissertation
Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Electro-acoustic technology and speech signal processing > Speech Signal Processing > Speech Recognition and equipment

Application and Research on Mathematical Morphology in Voice Anti-deliberate Imitation

Author WangRunDuo
Tutor ZhouPing
School Guilin University of Electronic Science and Technology
Course Applied Computer Technology
Keywords deliberate imitation speaker recognition mathematical morphology weighted morphological filter speech difference
CLC TN912.34
Type Master's thesis
Year 2011
Downloads 18
Quotes 0
Download Dissertation

Although the voice of people has personality, with the rapid development of modern computer technology, people’s voice can also be deliberately imitated. While the similarity of imitation speech is higher, current identity verification system may be deceived by deliberate imitator which has created a significant threat to information security on countries and individuals. Therefore, the researches of voice against the deliberate intent to imitate and the speech feature analysis of deliberate mimic speaker’s are launched. Thus it is very significant to find out the differences of the voice parameters between deliberate imitator and the original speaker, which has a wide range of applications in judicial expertise, criminal investigation, information security, etc.Firstly, this paper introduces the application and research status of the voice recognition technology against deliberate imitation, as well as the concept of deliberated imitation. The factors that affect the imitated speech are also pointed out, meanwhile it illustrates the harm to people on the daily lives caused by the deliberate imitation of speaker’s voice. Research of voice anti-deliberate imitation belongs to the areas of speaker recognition generally. The basic concepts of speaker recognition, principles and current research status are introduced in the text, and the commonly used methods of speaker recognition are also briefly described.Secondly, it is necessary to make the deliberate imitator voice and the voice deliberately imitated in noise-free status in order to achieve better results of voice anti-deliberate imitation. As the various environmental factors of actual lives, it is impossible to gain the ideal clean speech. But the effective methods can be taken to denoise the noisy speech as close as possible effects of the pure voice. As the voice signal is a complex nonlinear, non-stationary signal and some signal denoising methods based on traditional linear and stationary means have many limitations. Morphological filter with better filtering performance based on mathematical morphology is a new nonlinear filter. This filter can effectively suppress noise in speech signal processing and maintain the characteristics of speech signal edge details. Therefore, this paper has improved the traditional morphological filter and a weighted morphological filter is designed which can realize voice signal filtering and denoising, so that the ability to distinguish the deliberate imitators can be effectively improved. In addition, deliberate imitation speech database is established to make experiments. Differences of voice feature parameters based on traditional method and mathematical morphological are calculated respectively. And the two results are compared with each other. Experiments show that the characteristic parameters of voice anti-deliberate imitation based on mathematical morphology have the better performance.Finally, the work content of this paper is summarized, and the inadequacies in the paper are pointed out correspondingly. Besides the direction of future research in this field is proposed.

Related Dissertations
More Dissertations