Voiceprint recognition system and method based on multi-scale multi-level model

Abstract

The invention relates to the technical field of voiceprint recognition, and provides a voiceprint recognition method based on a multi-scale and multi-level model in order to improve the accuracy of voiceprint recognition, comprising the following steps: step 1, acquiring voice data with speaker labels; step 2, dividing the voice data into feature segments according to different scales, wherein each division scale corresponds to a hierarchy; step 3, constructing and training a multi-scale multi-level model corresponding to the data division mode; and 4, inputting to-be-recognized voice data into the multi-scale multi-level model obtained by training in the step 3 for voiceprint recognition. The invention discloses a voiceprint recognition system based on a multi-scale and multi-level model. The voiceprint recognition system comprises a data acquisition unit, a data division unit, a model construction unit, a model training unit and a voiceprint recognition unit. By adopting the mode, the accuracy of the voiceprint recognition model is improved.

Description

technical field [0001] The invention relates to the technical field of voiceprint recognition, in particular to a voiceprint recognition system and method based on a multi-scale and multi-level model. Background technique [0002] With the rapid development of artificial intelligence technology, more and more products incorporating artificial intelligence technology appear in people's daily life. Among them, voiceprint information, as an important biometric feature, is one of the effective ways for user identity verification. The mining and recognition of voiceprint information has also achieved good development and wide application in recent years, especially in the field of security and smart device products. [0003] However, the identity information contained in a piece of voice data is not evenly distributed on the voice signal, that is, different positions of the same voice signal show different distinctions for speakers. Therefore, the method of giving equal importa...

AI Technical Summary

Problems solved by technology

[0005] 1. Only the weight of each frame can be considered fixedly, but the discrimination of a continuous speech signal sub-interval as a whole for the speaker's identity cannot be considered in combination;

[0006] 2. It is difficult to solve the system performance degradation caused by the large difference between the length of the speech signal of the training corpus and the length of the test speech signal;

Images