A voice emotion recognition method for the elderly based on multi-feature fusion based on prediction

Abstract

The invention discloses a prediction-based multi-feature fusion speech emotion recognition method for the elderly, which comprises the following steps: obtaining an empty-nest elderly speech emotion database; extracting three different characteristic parameters for each speech and expression in the database; The prediction method fuses multiple features; uses SVM for feature recognition; outputs the emotion category with the highest accuracy predicted under the same segment of speech, and obtains the recognition result. The prediction-based fusion framework consists of two parts: the cross-feature prediction component, which combines three features by modeling the relationship between multiple features, and the connection of the three features is replaced by the first set of predictors, which learn the three features of the speech emotion category respectively. The mapping between feature parameters; the feature intra-prediction component models the time evolution of the three features respectively, and the feature intra-prediction component corresponds to the decision-level fusion. Each feature is modeled by two second sets of predictors, respectively Learn the mapping between past and current features for each category.

Description

technical field [0001] The invention belongs to the field of signal processing and pattern recognition, and more specifically relates to a voice emotion recognition method for the elderly based on prediction and multi-feature fusion. Background technique [0002] In recent decades, human-computer interaction technology has developed rapidly, but with the continuous emergence of various intelligent machines, people have begun to pay attention to this question: whether to let computers perceive emotions. As we all know, people will have ups and downs of emotions such as joy, anger, sorrow, and joy anytime and anywhere. When people make decisions or deal with things, too many emotional factors will lead to negative results. unattainable. But can a machine without emotion, unable to perceive emotion, play by ear? [0003] The term "affective computing" was first proposed by Professor Picard of the Massachusetts Institute of Technology in the book "Affective Computing" publishe...

AI Technical Summary

Problems solved by technology

[0005] In the prior art, when humans and computers communicate emotionally, it is impossible to guarantee that the accuracy of each emotion recognition rate has a high percentage, and various algorithms have obvious differences in the ability to express and distinguish different emotional feature vectors ;Speech recognition system has a strong dependence on environmental conditions and poor adaptability; it is easy to cause pronunciation distortion when used in a noisy environment; more than half of the recognition errors come from the endpoint detector; the selection of speech recognition primitives is not appropriate

Images