Research On Music Information Retrieval Algorithm Based On Deep Learning

The number of multimedia audio in network is increasing day by day.How to retrieve the desired target efficiently has become a key issue in music information retrieval technology.As a sub-task of music information retrieval,song humming matching has been paid more and more attention by people.In recent years,the rise of deep learning method has promoted the development of artificial intelligence and a new idea for music retrieval.As a new technology,deep learning has achieved a series of successes in speech signal processing and other fields.This paper draws on the research achievements of deep learning in speech signal processing.On the basis of combination of music information retrieval and deep learning theory,the main contents of research on how to make better use of deep learning to study music information retrieval are as follows:1.To extract the pitch and length of sequence information from standard WAVE Music Library.After considering the accuracy and time complexity of the algorithm,the new algorithm BP neural network is used to distinguish the music signal notes and clear the voiced sounds.On this basis,the pitch analysis is used to estimate the pitch period.2.The HMM model based on notes is established,the model training and recognition are carried out,and the music information retrieval based on HMM algorithn is realized.3.On the basis of Restricted Boltzmann Machine,the Convolution Deep Belief Network Model is introduced,and the algorithm is applied to this paper.It is pre-trained by unsupervised greedy layer by layer algorithm.the network parameters are fine-tuned by the supervised network training method,and the recognition ability of the model is improved by adjusting the network parameters.The experiment shows that the advantage of system recognition is more obvious when the song sample length is 3 s,and the recognition of model layer number CDBN2 is better than CDBNI.When the melody is accurate,the recognition rate of CDBN network retrieval is 4%higher than that of HMM.It is proved that CDBN can achieve better results in music retrieval and deep learning has great research value in the field of music retrieval.