Research On Retrieval Techniques Of MP3 By Humming

Along with the development of digital technology, MP3, with the high compression ratio and minimal distortion, becomes the most popular compression format of digital music, and widely spread on Internet. At present, the method of searching songs we want from huge quantity of MP3 sources, we can only search the song or singer's name to get this purpose. If only remember the melody, we will not retrieve the songs from the MP3 sources. The purpose of this paper is to research query by humming in MP3 retrieval, so, the user only need to hum part of a song from microphone and can get the song he or she wanted.Previous work in melody retrieval by humming has mainly focused on the MIDI type retrieval. Little attentions are put on the MP3 type. In addition, the humming way was demanded used special way, and the researches on humming in continuous way and lyrics way are quite less. Even if the system uses these humming ways, the melody track information uses the character representation and approximates character matching. In the aspect of extraction from humming signal, most systems use the traditional method to extract pitch; there are some problems, which the accuracy of pitch value will be influenced. Because the melodic tracks have a lot of data, it needs to design a quick melody-matching algorithm to match.Based on such shortcomings in studies on query by humming, this paper did some work as followed: In the melody extraction module, because the traditional algorithms have some shortcomings to extract the exact foundation frequency, the paper proposed a method, which combines the wavelet analysis with the autocorrelation function, to extract the pitch from the humming. In the module of building a melody database, firstly, the human voice must be extracted from the music; the process is called MP3 preprocessing. During the decoding of the extracted human voice, the method calculates the foundation frequency. An interval representation has been proposed and is used to represent the MP3 melody contour. The MP3 melody features database has been established based on the melody contour using the interval representation. In the melody-matching module, the paper designed a method of numeric indexing, established the indexes for the melody features database using this method, proposed a retrieval method of Dynamic Time Warping algorithm based on numeric indexing. Finally, we designed a system of query by humming in MP3 retrieval, analyzed the result of various modules' experiment, and proved effectiveness of the proposed methods.