Music genre classification method based on recurrent neural network and attention mechanism

Abstract

The invention discloses a music genre classification method based on a recurrent neural network and an attention mechanism. The method comprises: firstly, musical signal being transformed by a short-time Fourier transform to obtain a sonagraph, using a two-way recurrent neural network to learn features according to the from the sonagraph, to obtain higher-level abstract features, and using a parallel attention model to learn from the sonagraph to obtain attention probability distribution corresponding to feature representation, the attention probability distribution being used to set differentweights of musical feature representation; then performing weighted average on the features according to the feature weights to obtain fused features; finally, classifying music genres according to the fused musical features. In the method, the parallel recurrent neural network and the attention model are used, feature learning is carried out automatically according to the musical signals, and reasonable weights are set by using the attention probability distribution as the feature, and the features are weighted and averaged and then are classified, so that accuracy of music genre classification is improved, and complexity and limitation of manual feature extraction are prevented.

Description

technical field [0001] The invention relates to the field of music retrieval, in particular to a music genre classification method based on a cyclic neural network and an attention mechanism. Background technique [0002] It is difficult to classify and manage massive music data manually. For users, it is also necessary to be able to quickly retrieve interested music in a music library with a huge amount of data. So music genre classification has become one of the popular research directions in the field of music information retrieval. [0003] Classification through the traditional classification framework, the process of feature extraction is complex and difficult to implement, requires more professional prior knowledge in this field, and the music features extracted by hand lack versatility and are only suitable for a single classification task. Therefore, the present invention uses the cyclic neural network to automatically realize feature learning and obtain feature r...

AI Technical Summary

Problems solved by technology

[0003] Classification through the traditional classification framework, the process of feature extraction is complicated and difficult to implement, requires relatively professional prior knowledge in this field, and the music features extracted by hand lack versatility and are only suitable for a single classification task

Images