Text-independent voiceprint recognition method

Abstract

The invention discloses a text-independent voiceprint recognition method. The text-independent voiceprint recognition method comprises three stages of voiceprint recognition model training, extractionembedding and decision-making scoring. The model training stage comprises the steps of (1) voice signal preprocessing; (2) voice frame-level operation; (3) statistical convergence layer summary frame-level outputting; (4) one-dimension convolution operation; and (5) full joint layer output speaker sorting. After model training is completed, extraction embedding is conducted before nonlinearization of a first layer of full joint layers. Finally, decision-making scoring is conducted through the cosine distance to decide acceptance or refusal. A neural network embedding technology and a convolutional neural network are combined, one-dimension convolution and a maximum value convergence layer are used for dimensionality reduction, the number of convolution layers is increased, then deep-layerfeature extraction is conducted, and thus the property of a model is improved; and the cosine distance is used as the scoring standard, and the process is quicker and simpler.

Description

technical field [0001] The invention relates to the technical field of voiceprint recognition, in particular to a text-independent voiceprint recognition method combining neural network embedding technology and convolutional neural network. Background technique [0002] Voiceprint refers to the sound wave spectrum that carries speech information in human speech. Like fingerprints, it has unique biological characteristics and has the function of identification. It is not only specific, but also relatively stable. The sound signal is a one-dimensional continuous signal. After it is discretized, the sound signal that our common computer can process can be obtained. [0003] A discrete sound signal that a computer can process. Voiceprint recognition (also known as speaker recognition) technology, like the fingerprint recognition technology widely used on smartphones, extracts voice features from the voice signal sent by the speaker, and uses this to authenticate the speaker. r...

AI Technical Summary

Problems solved by technology

We know that the more layers of the network, the stronger the expressive ability, but it is very difficult to train the deep fully connected neural network through the gradient descent method, because the gradient of the fully connected neural network is difficult to pass more than 3 layers

Therefore, it is impossible for us to get a deep fully connected neural network, which limits its ability

Images