The Contribution of Music Information Retrieval System Optimization to Music Analysis in the Context of Big Data

Abstract

Abstract With the rapid popularization of Internet big data worldwide, people are able to transmit, download, and listen to huge amounts of music, which directly contributes to the demand for music information retrieval. In this paper, a music information retrieval system is constructed based on extracting music features. Both time and frequency domains characterize the music, and the transformation relationship between time domain, frequency domain, cepstrum domain, and power spectrum is proposed to extract music features. Further, the convolutional deep confidence network algorithm is applied to music information retrieval, an unsupervised greedy layer-by-layer algorithm carries out pre-training, and the network parameters are adjusted to improve the retrieval and recognition ability of the model. Functional validation of the system in this paper. In the music feature extraction experiments in this paper, the system’s accuracy for extracting feature points from different songs is more than 80%. In the music information retrieval experiments in nine different styles of music in the style of music in this paper, the average judgment of the system correct rate of 92.59%, in different proportions of the number of tracks in the retrieval success rate, is higher than 88%. In music analysis fields such as music recommendation and music soundtrack design, the music information retrieval system constructed in this paper plays a significant role.