Method and system for improving degree of accuracy of voiceprint recognition based on recognition models

Abstract

The invention discloses a method and system for improving the degree of accuracy of voiceprint recognition based on recognition models. The method includes the following steps: obtaining voiceprints of a same user under different states to build a plurality of background recognition models which include a plurality of background voiceprint feature vectors; building standard recognition models according to the background recognition models to obtain unrecognized voiceprint features in unrecognized voiceprints, the unrecognized voiceprint features including wavelet elements of the unrecognized voiceprints; processing according to the unrecognized voiceprint features and standard voiceprint features in voiceprint recognition models to obtain the distinguishing degrees of the unrecognized voiceprints; judging whether the distinguishing degrees are larger than a preset standard threshold value, and when the distinguishing degrees are larger than the preset standard threshold value, recognizing the unrecognized voiceprints as locked voiceprints; generating recognition records for the standard recognition models; and making statistics of recognition records of the standard recognition models, forming a statistical result and then performing descending sorting on the standard recognition models. The beneficial effects of the technical scheme are that accurate recognition of unrecognized voiceprints by the user under different states is satisfied, and rapid recognition of the unrecognized voiceprints can be realized.

Description

technical field [0001] The present invention relates to the technical field of voiceprint recognition, in particular, to a method and system for improving the accuracy of voiceprint recognition based on a recognition model. Background technique [0002] Voiceprint recognition technology, also known as speaker recognition technology (Speaker Recognition), is considered to be the most natural biometric authentication technology. , reflect the size and shape of the mouth and the spectral characteristics of the length of the vocal tract, etc.), and then identify the speaker identity and other work technologies, which are being used by more and more smart devices. The specific voiceprint recognition (Voiceprint Recognition, VPR) , which includes two categories, namely: 1) Speaker Identification and 2) Speaker Verification. The former is used to determine which one of several people speaks a certain speech; the latter is used to confirm whether a certain speech is spoken by a des...

AI Technical Summary

Problems solved by technology

[0003] However, the existing voiceprint recognition technology is not very perfect. The recognition model established by the user is often affected by other factors, such as the voice of the same person is easily affected by physical condition, age, emotion, etc., or different microphones and Channels have different effects on voiceprint recognition, or environmental noise interferes with voiceprint recognition, which will affect the accuracy of voiceprint recognition

Images