Construction of Music Intelligent Creation Model Based on Convolutional Neural Network

Area of Science:

• Artificial Intelligence
• Music Technology
• Signal Processing

Background:

• Current intelligent music creation methods struggle with weak feature expression due to fixed coding steps in audio data.
• Deep learning offers potential for enhanced feature representation in complex audio signals.

Purpose of the Study:

• To propose a novel deep music intelligent creation method based on convolutional neural network theory.
• To address the limitations of high data feature dimension and poor recognition performance in existing methods.

Main Methods:

• A convolutional recurrent neural network was employed to generate effective hash codes.
• Music signals were preprocessed into Mel spectrograms and fed into a pretrained Convolutional Neural Network (CNN).
• Feature maps from CNN layers were selected to construct a feature map sequence, reducing data dimensionality.

Main Results:

• The proposed method effectively reduced high-dimensional music data dimensionality.
• The intelligent music creation correct rate reached as high as 98%.
• Characteristic signal distortion rate was reduced to below 5%.

Conclusions:

• The deep learning approach significantly improves algorithm performance for intelligent music creation.
• The method enhances the feature expression ability and recognition accuracy in music generation.
• This research provides an effective solution for intelligent music creation with complex audio data.