METHOD OF RECOGNIZING SPEECH EMOTIONS USING 3D CONVOLUTIONAL NEURAL NETWORK Russian patent published in 2024 - IPC G06N3/08 G10L15/08 

FIELD: physics.

SUBSTANCE: present invention relates to computer-implemented methods and systems for assessing emotions and, more specifically, to methods for recognizing speech emotions using 3D convolutional neural network. Method of recognizing speech emotions using three-dimensional (3D) convolutional neural network, implemented using a computer and comprising two main steps, at the first stage three-dimensional tensors are provided using the reconstructed phase space of the input speech signals, and at the second step, using 3D convolutional neural network, trained based on the three-dimensional tensors and the corresponding emotion labels, performing the prediction of the emotions contained in the input speech signals, wherein the reconstructed phase space is used as follows: a one-dimensional signal is displayed in a three-dimensional space, and then the three-dimensional tensor is extracted for use as input data of 3D convolutional neural network.

EFFECT: high efficiency of recognizing speech emotions using 3D convolutional neural network.

4 cl, 6 dwg