A New Musical Features Extraction Algorithm Based On MIDI Files And FPGA Design

As a carrior of the art of emotion expression,music has a long history.With the development of material and cultural life,as well as the development of science and technology,people are no longer confined to use the ears to listen to the music.They also want to feel the rhythm of the music in many other ways.The challenge is posed by the extraction of the characteristics of composition and music,so we need a new approach to create and interpret the music,which involve the extraction of musical features.However,how to extract the theme quickly and efficiently has always been a challenge.This is also the main contents of the paper.MIDI as one of the most popular digital music format,is easy to be editted and has small file size.It also has many other characteristics,as well as a wide range of applications.Based on the above background,we extract musical features from MIDI files for research and make the circuit design for a follow-up.This dissertation requires the knowledge of the basic music theory,so we introduce the format of MIDI files at first.Through the analysis of meta in the header,we can get some MIDI files,which have already been marked main melody tracks,as well as information such as tempo.This is useful for the validation of melody extraction algorithm.In MIDI files,by each State represented by a hexadecimal number bytes and bytes of data analysis,we can get the pitch of each note,sound,tone,and so on.We calculate the variable-length encoding delta-time and extract the starting time and duration of the notes to translate the MIDI files to the key matrix.In melody extraction method,we improve the common algorithm,skyline.We remove tenth channel drum tracks and the notes which have too short during time after the skyline algorithm.We also remove the notes whose during time is 1/3 long than the average time and improve recognition accuracy.Base on this,we use clustering method to categorize similar tracks and select main melody from each class as audio tracks.Then we simulate the algorithm by MATLAB,and verifies the accuracy of algorithms and the main melody extraction.Then make hard wire design for the skyline algorithm based on FPGA and sucessfully implement the extraction of main melody in MIDI music files.And make board level debugging.The results of circuit were compared with results of MATLAB to verify the correctness of the circuit.