Method and device for verifying vocal print irrelevant to text and computer device

Abstract

The invention discloses a method and device for verifying a vocal print irrelevant to a text and a computer device. The method includes the steps of intercepting first voice with frequency lower thandesignated frequency in to-be-verified voice, extracting frequency spectrum features of the first voice through a Mel-frequency cepstral coefficient, extracting first vocal print features of the frequency spectrum features through a preset vocal print feature extracting model based on deep neutral network training, finding second vocal print features matched with the first vocal print features ina preset vocal print database, and if the second vocal print features matched with the first vocal print features are found, judging that the to-be-verified voice passes verification. By extracting the vocal print features of the to-be-verified voice through the preset vocal print feature extracting model based on the deep neutral network training, the method has higher extracting efficiency, a memory, a video memory and the like of a system are saved, the vocal print feature extracting model can continuously conduct self-learning, and the vocal print feature extracting accuracy can be kept consistent.

Description

technical field [0001] This application relates to the computer field, in particular to a text-independent voiceprint verification method, device and computer equipment. Background technique [0002] The text-independent voiceprint system mainly uses the different speech acoustic features (pronunciation organ features and behavioral features) between individuals to distinguish speakers. The core of the currently widely used text-independent voiceprint system has two parts: the acoustic feature i-vector is extracted through the Gaussian mixture model-universal background model (GMM-UBM); the acoustic feature i-vector is extracted through the probability linear discriminant (PLDA). similarity score. However, there are the following disadvantages: (1) GMM-UBM model performance is difficult to optimize; (2) there are many calculation steps; (3) GMM-UBM model has many parameters, and model engineering takes up a lot of memory and video memory. Therefore, providing a new text-in...

AI Technical Summary

Problems solved by technology

[0003] The main purpose of this application is to provide a text-independent voiceprint verification method, device and computer equipment, aiming to solve the problem of the many calculation steps for voiceprint feature extraction in the prior art, the difficulty in optimizing the performance of the GMM-UBM model, and the relatively large parameters of the GMM-UBM model. Many, model engineering takes up memory, video memory is large, etc.

Images