Intelligent Music Chord Recognition and Evaluation Based on Convolution and Attention

How to use objective evaluation for intelligent aesthetic assessment of music has been an interesting topic for musicologists and scientists. As the skeleton of music, chords have an indispensable importance in the objective evaluation of music. Therefore, in this paper, a convolutional neural network equipped with self-attention mechanism is used to realize the chord classification task, and Learnable Classifier is trained to improve the classification accuracy of rare chord classes, and the classification accuracy and confusion matrix in the mixed dataset of UsPop2002 and POP909 also proves that the model proposed in this paper has good classification accuracy. Meanwhile, this paper optimizes the quantitative indexes of chord entropy, rhythmic stability and chord progression consistency to achieve the task of chord aesthetics calculation, and also adopts the form of circle of fifths to visualize the main chords. Although the current music evaluation is still subjective, the proposed intelligent chord analysis and evaluation model provides a possibility for music evaluation.