Data processing device and data processing method

Abstract

The invention relates to a data processing device and a data processing method. The data processing device comprises an extraction unit, a division unit, a calculation unit and a training unit, wherein the extraction unit is used for extracting an i vector of training audio data from each of the multiple pieces of training audio data; the division unit is used for dividing the i vectors into multiple clusters and calculating the cluster center of each cluster; the calculation unit is used for calculating the distance between the i vector of each piece of training audio data and the cluster center of each cluster; and the training unit is used for training a deep neural network (DNN) model, and the training unit enables the distance between the i vector of each piece of training audio dataand the cluster center of each cluster to serve as an output true value of the DNN model. The DNN model can be trained by using the data processing device and the data processing method of the invention so as to output the distance between the i vector of the audio data and each cluster center, so that the amount of calculation generated in the audio data registration and recognition process is reduced, and more sufficient label information can be obtained at the same time.

Description

technical field [0001] Embodiments of the present invention relate to the field of data processing, in particular to a data processing device and method capable of training a deep neural network DNN model, a data processing device and method capable of registering audio data, and a data processing device and method capable of testing audio data. Background technique [0002] This section provides background information related to the present disclosure which is not necessarily prior art. [0003] Speaker recognition is a biometric recognition technology, also known as voiceprint recognition. Traditional speaker recognition technology mainly includes two types: one is based on GMM (Gaussian Mixture Model, Gaussian Mixture Model) to extract i-vector (also known as i-vector or identity vector), register and process audio data according to i-vector Recognition; the other is to extract the d-vector (d-vector) based on DNN (Deep Neural Network, deep neural network), and register ...

AI Technical Summary

Problems solved by technology

In the recognition technology based on GMM to extract the i-vector, it is necessary to extract the super-vector of the audio data first, and the super-vector can obtain the i-vector of the audio data after eight matrix operations and one matrix inversion operation. The algorithm is complex and time-consuming; in addition , if the amount of data used to train GMM is reduced, the accuracy of recognition will drop significantly

In the recognition technology based on DNN to extract the d vector, due to structural limitations, the output layer during training has to be abandoned, and the data of the last hidden layer is used as the d vector; further, there are fixed output nodes in this system When the training set is updated, the DNN model needs to be retrained; in addition, only speaker information is used as a label in this recognition technique, and a large amount of information such as channel, sentence content and noise are discarded

Images