Voiceprint recognition model training method, device and computer system

Abstract

The invention discloses a voiceprint recognition model training method and device based on a federated learning technology, and a computer system. The method comprises the steps: enabling a federatedlearning terminal to carry out distortion processing on an original audio signal, wherein the distortion processing is used for carrying out distortion processing on the original audio signal in a time domain or masking processing on a frequency domain; performing acoustic feature extraction on the audio signal; splicing the randomly generated data number with the extracted corresponding acousticfeatures to obtain information vectors; performing encryption processing of lossless mathematical features on the spliced information vectors to obtain to-be-trained data, and transmitting the to-be-trained data to a federated learning server; receiving the to-be-trained data by the federated learning server; splitting a plurality of information vectors in the to-be-trained data into data numbersand acoustic features in a one-to-one correspondence manner according to decryption rules; and training an acoustic vector extraction algorithm by taking the acoustic features as input and the data numbers as output, and completing training of the acoustic vector extraction algorithm under the condition that the acoustic vector extraction algorithm is converged.

Description

technical field [0001] The present application belongs to the field of voiceprint recognition, and in particular relates to a voiceprint recognition model training method, device and computer system. Background technique [0002] In the era of big data, artificial intelligence practitioners and data collectors belong to different divisions of labor. The two need each other's resources, but it is often difficult to build trust and work together due to security, privacy and other issues. To be precise, data collectors need to hand over their data use rights to artificial intelligence practitioners, allowing the latter to use data to train artificial intelligence models, so that their data can generate value; artificial intelligence practitioners need to obtain data collection The right to use the data provided by the party can train the applicable model and transfer the right to use the model, thereby further promoting the development of society. In the process of implementin...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the embodiment of the present application is to provide a voiceprint recognition model training method, device and computer system, which can solve the current technical problem that it is difficult to ensure that the data provided by the data collector can be safely and effectively used by artificial intelligence practitioners

Images