A Speech Emotion Recognition Method Fused with Long Span Emotion History

Abstract

The invention discloses a speech emotion recognition method which integrates long-span emotion history. The method comprises the following steps: using different parameters in the time domain and transform domain to perform endpoint detection, eliminating non-speech data in the original speech sequence, and obtaining speech segment data to be recognized; dividing the speech segment data to be recognized into independent speech segments Data unit; use the first support vector machine to perform preliminary emotional state classification on the speech segment data unit respectively; add window to the preliminary classification result of the emotional state, and use the second support vector machine to perform fusion to obtain emotional recognition with long-span emotional history result. While ensuring high-precision classification of local units of speech signals, the present invention makes full use of context information in long spans of signal sequences to achieve optimal classification results for each unit in the sequence. The invention can be used for emotion recognition of speech signals, and has the advantages of good real-time performance, greatly improved recognition accuracy and the like.

Description

technical field [0001] The invention belongs to the field of speech signal processing, and in particular relates to a speech emotion recognition method which integrates long-span emotion history, and thereby improves the accuracy of continuous speech emotion recognition. Background technique [0002] For decades, researchers at home and abroad have done a lot of research work on speech emotion recognition and proposed many effective algorithms for emotion recognition. These methods can be divided into detection methods based on static classifiers and detection methods based on dynamic classifiers in terms of processing strategies. Detection methods based on static classifiers mostly use support vector machines (SVM), neural networks, Boosting, etc., and most of these classifiers are discriminative models. Due to its strong discrimination ability, it is widely used in the field of emotional state recognition, but this method ignores the interconnection between the emotional ...

AI Technical Summary

Problems solved by technology

However, since HMM is a generative classification model, its classification performance is not as good as that of discriminative classifiers.

At the same time, it cannot model long-span contextual information, that is, the fusion range of emotional history is limited, and it cannot fully reflect the key role of emotional history information in emotion recognition.

Images