Speaker recognition method based on sound-induced electroencephalogram signals

Abstract

The invention discloses a speaker recognition method based on sound-induced electroencephalogram signals. According to the method, by collecting electroencephalogram data, fusion features of time-frequency features and time-domain statistical features of an auditory stimulation part are extracted; fusion features obtained by electroencephalogram signals of an alpha frequency band baseline correction part are used as a background template; and the background template is subtracted from the auditory stimulation part fusion feature to obtain a clean task state data fusion feature, and finally different speakers are distinguished by using the network model provided by the invention. The invention provides a feasible speaker recognition method based on the sound-induced electroencephalogram signals, different speakers are distinguished by using the trained classifier, and the accuracy rate reaches 90%.

Description

technical field [0001] The invention belongs to the field of speaker recognition based on EEG signals in the technical field of EEG change analysis, and in particular relates to a method for distinguishing different speakers based on the analysis of EEG signals induced by sound. Background technique [0002] Traditional speech-based speaker recognition has been very mature, and we can extract individual differences from the speech of different speakers. Acoustic features such as spectrum, cepstrum, and formant, etc., as low-level features, can represent differences in the vocal tract structure of different speakers. Phonological features such as prosody, rhythm, dialect, etc., as high-level features, can represent the differences in speaking styles of different speakers. [0003] Individual differences among different speakers will lead to differences in the EEG signals evoked from the subjects. This difference between speakers is mainly reflected in three aspects. The fir...

AI Technical Summary

Problems solved by technology

For example, most of the hearing aids on the market now have a shortcoming - they cannot accurately amplify the voice of the speaker that the patient is concerned about, because these hearing aids have a predefined assumption - to amplify the voice of the speaker directly in front of the patient, this fixed The mode is very unfriendly to the patient. If the target speaker that the patient pays attention to can be detected in real time, and then the voice of the target speaker is amplified, and the voice of other people is suppressed, the performance of the hearing aid will be greatly improved.

Images