| With the development of big data,a large number of songs have emerged rapidly.The traditional music analysis method,which relies on artificial labels of experts,has gradually been unable to adapt to the background of rapid updating of songs.Considering the domain knowledge needed by music analysis,the problems exist in the process of automatic music analysis are as follows.Firstly,as for music expression,the symbolic representation of audio signal for music score type can help to analyze the melody of music more intuitively.Due to the complexity of music signal components,such as the mixture of voice and accompaniment,the diverse types of instruments,the variety of chord composition,the extraction of music score is very difficult.Secondly,as for music analysis,the analysis of various types of music data including audio,music score and lyrics can help us understand music more comprehensively.How to construct an analysis model that integrates multiple types of data will be helpful for music similarity analysis and calculation.Thirdly,as for music retrieval,the mainstream retrieval methods need to rely on annotated text information,while the audio content-based retrieval methods rely on feature extraction and template matching.It is urgent to improve the efficiency and innovate the mode for music retrieval.In view of the above problems,this paper carries out the research and implementation of music analysis and retrieval platform based on music score generation.Firstly,in music representation,a method of music score generation based on audio signal is proposed,which converts audio signal into symbolic music score representation.Secondly,in music analysis,a music analysis model based on music score data combined with multiple data types is proposed,and on this basis,the similarity calculation method is studied.Finally,in music retrieval,content-based retrieval method is constructed by combining music analysis model.The main contents of this paper are as follows:(1)The overall research framework of music analysis and retrieval platform based on music score generation is proposed.This paper proposes a general research framework of music analysis and retrieval platform based on music score generation,aiming at music application on the basis of automatic analysis of music content.The framework is based on the construction of music data processing process.It focuses on music data acquisition layer,processing layer and application layer to study the relevant methods and solve the problems in the process of music analysis and retrieval.(2)An automatic music score generation method based on audio signal analysis is proposed.This paper presents an automatic music score generation method based on audio signal analysis.Methods in view of the complexity of music audio components,the problems of separation of human voice and accompaniment,single tone recognition,chord recognition,and note sequence extraction are studied.(3)The music analysis model integrating music score and audio lyrics is constructed.Based on the method of music score generation,this paper constructs a music analysis model.Considering the diversity of music data types,including audio,music score,lyrics and so on,the model fuses various data types,and studies the method of model similarity analysis.(4)Music retrieval based on music analysis model.Combining with music analysis model,this paper carries out music retrieval application based on music content.In order to improve the existing music retrieval application mode,a music retrieval method based on musical sequence is proposed.The efficiency of music retrieval can be improved by similarity calculation of music analysis model.(5)A prototype system of music analysis and retrieval platform is constructed.According to the methodological framework and research methods proposed in this paper,a prototype system of music analysis and retrieval platform is constructed,which provides music score generation,music analysis,music retrieval and other services.With the development of prototype system,the application scenario of this method is introduced.At the same time,the platform is compared with existing platforms,which proves the effectiveness and practicability of our method. |
