Font Size: a A A

Research And Development Of Voiceprint Recognition Based On Machine Learning

Posted on:2020-06-24Degree:MasterType:Thesis
Country:ChinaCandidate:X B QiFull Text:PDF
GTID:2438330575960146Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
Sound is the information carrier for communication between people,and sound plays a decisive role in human-computer interaction.Voiceprint recognition is a very important direction in speech recognition.The application of this technology to human-computer interaction will greatly improve the security of human-computer voice interaction.As a biometric authentication method,voiceprint recognition has many important application prospects.In recent years,machine learning technology has made major breakthroughs in the field of automatic speech recognition.More and more machine learning methods,especially deep learning methods,have been introduced into voiceprint recognition and have achieved remarkable results.The i-vector based voiceprint recognition method is currently the benchmark method for text-independent voiceprint recognition.However,this method has a low recognition rate in the case of short-term speech and is also susceptible to noise interference.This paper uses the theory of machine learning to design a voiceprint recognition method based on time-delay neural network.Compared with the benchmark method,this method improves the recognition rate and stability of the voiceprint system,especially in the short-term sound recognition effect.In order to further improve the recognition effect of the system,this paper designs a voiceprint recognition method based on generation vector,which combines the voiceprint recognition method based on i-vector with the voiceprint recognition method based on time delay neural network..This method uses typical correlation analysis to fuse part of the i-vector information into the feature vector extracted by the delay neural network,so that the generated vector can better represent the identity characteristics of the speaker.The three voiceprint recognition methods are experimentally verified.The equal error rate EER of the generated feature vector model is reduced by 3.1% compared with the traditional i-vector model.In this paper,the actual application test of the designed model is carried out to verify the feasibility of the system.
Keywords/Search Tags:voiceprint recognition, feature extraction, machine learning, i-vector
PDF Full Text Request
Related items