Speaker recognition method and system based on unsupervised scene adaptation

Abstract

The invention discloses a speaker recognition method and system based on unsupervised scene adaptation. The method comprises the following steps of: acquiring a source domain training set and a target domain training set; preprocessing the source domain training set and the target domain training set to obtain a preprocessed source domain training set and a preprocessed target domain training set; training and optimizing a collaborative adversarial network model by using the preprocessed source domain training set, the preprocessed target domain training set and a loss function; inputting the voices of a plurality of speakers and the identities of the speakers corresponding to the voices of the speakers into the optimized collaborative adversarial network model, and constructing a speaker voice database; and inputting the voice of a speaker to be recognized into the optimized collaborative adversarial network model, and recognizing the identity of the speaker corresponding to the voice of the speaker to be recognized by using the optimized collaborative adversarial network model and the speaker voice database. According to the method, the speaker recognition capability is not lost when the problem of scene mismatching of speaker recognition is solved, so that the speaker recognition precision is improved.

Description

technical field [0001] The present invention relates to the technical field of speaker recognition, in particular to a speaker recognition method and system based on unsupervised scene adaptation. Background technique [0002] Voiceprint authentication is based on the diversity and specificity of voice. It also has the advantages of non-contact authentication, convenient data collection, and suitable for remote authentication. Voiceprint authentication, together with biometric authentication such as fingerprints and faces, has become an important method in the field of biometric authentication. . Speaker recognition technology is the core of voiceprint authentication, and its performance directly affects the authentication accuracy of voiceprint authentication. When dealing with voiceprint authentication scenarios, it is often encountered that the registration voice and the test voice scene are different, such as registration in a quiet environment and testing in a noisy en...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a speaker recognition method and system based on unsupervised scene adaptation, which can improve the accuracy of speaker recognition without losing speaker recognition ability when solving the scene mismatch problem of speaker recognition

Images