Adaptive Music Feature Decomposition Classification Model Based on CEEDAM and Deep Learning

Abstract. Accurate music genre classification is very important in applications such as music recommendation systems, digital libraries, and music streaming services. A classification feature with more fine-grainedness can more accurately reflect the characteristics of music.This study proposes a new method for music feature acquisition for music genre classification by integrating the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise method and deep learning technology. By decomposing the audio signal into intrinsic mode functions, whcih called CEEMDAN, and extracting meaningful features from the mel-spectrogram using Pre-trained models, the proposed model effectively captures a variety of more fine-grained frequency information necessary to distinguish music genres. At the same time, The addition of the attention mechanism can further enhance the ability of our model to focus on the most relevant features that best reflect the underlying information of the music.This music feature classification method will provide a reference perspective for music signal processing.