A Speaker Recognition Method Based on Mutual Information Estimation

Abstract

The invention discloses a speaker identification method based on mutual information estimation, which solves the problems of poor distinguishability of speaker identity features and high error rate of the identification system. During training, the spectrogram is first extracted from the speech and used as the input of the VGG-M network; then random triplet sampling is performed on the training data, positive and negative samples are obtained for mutual information estimation, and the The objective function trains the network. During recognition, the trained VGG‑M network is used to extract the embedded features corresponding to the test voice and the target speaker voice; then the cosine distance between the two embedded features is calculated and used as the speaker matching score; the score is compared with The set threshold is compared to determine whether the test speech comes from the target speaker. This method can effectively utilize the mutual information between the speaker features corresponding to the positive and negative samples, so as to optimize the network training and reduce the error rate of the system. The present invention can be applied to the field of speaker recognition.

Description

technical field [0001] The invention belongs to the technical field of speaker identification, in particular to a speaker identification method based on mutual information estimation. Background technique [0002] In recent years, biometric identification technology has gradually become a convenient and quick way to verify identity information. Voice is the most commonly used and direct way of communication, and the unique physiological characteristics of each person obtained from voice are called "voiceprint". Due to the individual differences in each person's vocal organs and pronunciation habits, each person's voiceprint is different and unique. Therefore, the unique biometric features of the speaker can be extracted from the speaker's speech signal as the uniquely authenticated identity information. [0003] With the rapid development of deep learning in image processing, speech recognition and other fields, methods based on deep learning are gradually being applied in...

AI Technical Summary

Problems solved by technology

However, when directly using the deep neural network for unsupervised learning and extracting speaker features, it is impossible to judge whether the speaker feature representation extracted by the network judgment is unique and has high expressive ability

Images