Music Emotion Recognition Based on a Neural Network with an Inception-GRU Residual Structure

As a key field in music information retrieval, emotion recognition is indeed challenging task. To enhance the accuracy of classification and recognition, this paper uses idea inception structure to use different receptive fields extract features dimensions perform compression, expansion, recompression operations mine more effective connect timing signals residual network GRU module features. A one-dimensional (1D) Convolutional Neural Network (CNN) with an improved Inception Gate Recurrent Unit (GRU) was presented tested on Soundtrack dataset. Fast Fourier Transform (FFT) used process samples experimentally determine their spectral characteristics. Compared shallow learning methods such as support vector machine random forest deep method based Visual Geometry Group (VGG) CNN proposed by Sarkar et al., 1D Inception-GRU demonstrated better performance tasks, achieving 84%.