Speech emotion feature extraction method based on transformer model encoder

Abstract

The invention discloses a speech emotion feature extraction method based on a transformer model encoder, and is suitable for the field of artificial intelligence and speech emotion recognition. The method comprises the following steps: firstly, extracting low-level speech emotion features from an original speech waveform by using a sincnet filter, and then further learning the low-level speech emotion features by using a multi-layer transformer model encoder; wherein the improved transformer model encoder is characterized in that a layer of sincnet filter is added in front of a conventional transformer model encoder, namely a group of parameterized sinc functions with band-pass filters, the low-level feature extraction work of original voice waveform signals is completed by utilizing the sincnet filter, a network is enabled to better capture important narrow-band emotion features, and therefore, the frame-level emotion features containing the global context information at a deeper level are obtained.

Description

technical field [0001] The invention relates to a speech emotion feature extraction method, in particular to a speech emotion feature extraction method based on a transformer model encoder used in the fields of artificial intelligence and speech emotion recognition. Background technique [0002] With the advancement of science and technology, human-computer interaction has become an important research field. Speech emotion recognition technology can make machines more human. At present, speech emotion recognition has been studied for more than ten years. Its essence is the simulation of the process of human emotion perception and understanding by computer. The task is to extract effective emotional acoustic features from preprocessed speech signals and find out these acoustic features. Mapping relationship with human emotion. [0003] In the research of speech emotion recognition, how to extract the features with the most emotional information from speech signals is still ...

AI Technical Summary

Problems solved by technology

However, there are still many systems that use neural networks such as RNN and DNN to extract learning features. However, due to the problem of long-distance gradient disappearance and information loss from long sequences to fixed-length vectors, the traditional neural network is not yet able to fully understand the global information of speech emotion. Failed to extract well, there are great limitations, and this property is very important in the emotional characteristics of speech

Images