Research And Realization Of A Web Based Query-by-humming System

In recent years, with the rapid development of multimedia and network technology, digital music products are rapidly increasing. Searching songs only by the name of the song, singers, lyrics and some text information, has been unable to meet the needs of people. Along with the amount of music constantly expanding, people are difficult to remember all the text information, facing the mass information, people may remember a melody, perhaps is a fragment. So how to fast, efficient query music by humming the melody becomes an urgent problem.Compared to the mature video and image retrieval, Music retrieval is lagging behind, but it gradually becomes a research focus in the field of modern multimedia retrieval.Content-based music retrieval manage music information in a new perspective, based on the rhythm of the music, melody, pitch. Query-by-humming as a branch of the content-based music retrieval, it can search the needed song by users humming.This paper studies a query-by-humming retrieval system based on Web, people can use the singing way as input, the system returns the needed song to the user after retrieval the song information in MIDI music library. The main work of this paper includes two aspects:come up with a certain anti-noise integrity of the pitch frequency extraction framework; and an improved RA (Recursive Alignment) matching algorithm.In this paper, the query-by-humming system retrieval experiments were conducted with821humming segment and the collection of1300MIDI music database, TOP-10retrieval accuracy rate is92.2%, the experiment proved the effectiveness of the pitch frequency extraction framework and the improved RA algorithm in this paper.