| Rapid development of modern information, multimedia and network technology has resulted in large and ever-increasing stores of multimedia data. Therefore, content-based audio information retrieval technology has been attracting more and more attention to the efforts to make full use of existing audio information. The research on content-based musical retrieval is composed of the study of musical theoretic, signal processing and pattern recognition. It is significant for the management of the music digital library, the music search in the Internet and daily amusement.Normally, the whole content-based musical retrieval system can be divided into two parts: musical feature retrieval and music searching. The latter is largely dependent on the former one. Thus, this article is mainly about musical feature retrieval techniques.Firstly, the article gives detailed descriptions on the whole system of content-based music retrieval and then the process of human voice feature retrieval. Furthermore, analysis on the content-based music retrieval techniques on both time-domain and frequency domain is studied in the article, together with three new methods of melody pitch retrieval. The first two time-domain algorithms focus on improving pitch retrieval accuracy rate based on mature algorithms. The first method makes use of musical scale, which can effectively retrieve the human voice in the musical background by combining musical scale and AMDF algorithm and by establishing the model on voice energy; the second method adopted an open-loop and close-loop framework based on human voice feature and SIFT algorithm; the third frequency-domain-based algorithm described a new pitch retrieval method by using some special features of frequency data during MP3 decoding process, and gives good results on Matlab testing platform;The study work of this thesis provides the preparation for the developing of content-based musical retrieval research and pushes forward the study on melody retrieval. |
PDF Full Text Request