Voice personality prediction method based on convolutional neural network

Abstract

The invention discloses a voice personality prediction method based on the convolutional neural network, comprising the following steps: S1, establishing a training sample set, including: converting voice data files into a spectrogram, performing big five personality questionnaire tests on acquisition subjects, calculating average value and standard deviation of personality dimension scores, and marking the personality dimension scores which meet the preset requirements; S2, training a convolutional neural network model, including: training the neural network model by the BP algorithm, calculating network errors and updating weights during iteration, training until accuracy of a validation is no longer increased and then stopping training, and saving neural network parameters as the neuralnetwork model; and S3, performing personality prediction on voice data to be tested by the utilization of the convolutional neural network model. Through the technical scheme, the disadvantage that atraditional method requires strong prior knowledge to assist extraction of features with distinguishing degree is overcome, and automatic extraction of phonetic features is realized.

Description

technical field [0001] The invention relates to the field of computer application technology, in particular to a voice personality prediction method based on a convolutional neural network. Background technique [0002] Personality is an individual's internal tendency in behavior, which represents a stable behavioral difference between people. Accurate assessment of personality can provide users with accurate personalized services in areas such as consumption and recruitment. Speech is a person's "auditory face" and there is a close connection between personality, and a person's personality traits can be inferred from his speaking style and pronunciation characteristics. Traditional personality assessments use personality questionnaires in the form of questions and answers. The steps are not only complicated but also must rely on the cooperation of users. [0003] Traditional methods need to extract some features of speech, including acoustic features and prosodic features...

AI Technical Summary

Problems solved by technology

These artificial features cannot fully reflect the information of speech, and the design of artificial features requires strong prior knowledge

These shortcomings determine that it is difficult to design the optimal distinguishing features in the actual system development of the existing technology, which affects the further improvement of the performance of the speech prediction personality model

Images