Musical Retrieval By Rhythm Humming

With the fast development of network and multimedia technology, an increasing large amount of multimedia information is available on the internet. As an important part of multimedia information, the amount of audio data increased dramatically as well. Thus, how to easily and rapidly find out the audio information that we want has become problems needed to be solved. In recent years, the query-by-humming technique is adopted by many researchers as an ideal means to solve the above problems.As one branch of the study of content-based musical retrieval, the query-by-humming technique touches musical methodology, signal processing and pattern recognition. And this technique plays a significant role in network music sharing and daily amusement.There are five aspects of research work mainly presented in this paper, which include:(1) According to the research on the musical signals analyzing theory, a method using the pitch difference and the note duration ratio to describe the musical melody features is presented. Comparing with the traditional method using pitch difference only, the method in this paper describes the melody features better. So, the result we get can be more accurate.(2) Based on summarizing of the advantages in extracting musical melody features from MIDI files, we confirm MIDI to be the database file format and present the algorithm to find out the main melody from the MIDI files.(3) Some new algorithms and implementing steps about humming music processing are described in details, which include the signal pre-processing, pitch extraction, features post-processing, note segmentation, etc. Meanwhile, these new algorithms take people's humming habits and the problem which will come up in the processing into account. So the result of humming music processing can be more accurate.(4) A new matching method based on edit distance matrix is presented and the similarity of pitch difference and note duration ratio between the humming segment and the database music is computed. At last, results are returned to the client in sequence of the similarity.(5) Factors that may affect the experiment result are investigated in details and the strong points and shortcomings of our system are carefully summarized. At last, our algorithm is compared with other different algorithms in aspects of the experiment environment, the retrieval time and the hit rate.The research on query-by-humming system is of great value in theory and practicality. People hope that computers can understand music, so the work in this paper carried out a useful exploration into issues involved in this field.