Based On Sound Recognition Of Polyphonic Music Of The Multi-label Classification

In recent years, With the fast booming of online music repositories , there are increasing needs for content-based Automatic Indexing to help users find their favorite music objects, Modern society ease the fast-paced by music, highly competitive state of mind and emotions under pressure, has become a trend.Music instrument recognition is one of the main subtasks. Because the same music with different instruments to play or ensemble, will give the audience a different feel and effect. Numerous successful approaches on musical data feature extraction and selection have already been proposed for instrument recognition in monophonic sounds. Unfortunately, none of those algorithms can be successfully applied to multi-timbral sounds. Thus, identification of music instruments in multi-timbral sounds is still difficult and challenging, especially when harmonic partials are overlapping with each other. From the audio file, identification and separation of sound from a variety of musical instruments is very useful. When people in the melody played by a specific instrument to search, its analysis of the results of the sound will automatically index and browse the music data is very helpful. This has stimulated the research on multi-labeled instrument classification and new features development for content-based automatic music information retrieval. In this dissertation,we study and explore the music retrieval based on hummed melody.In this paper, we propose the background of music and studying the status and goals, refer to timbre features based on MPEG-7, and increase the time attribute, generate a new time-domain features. The experimental results show that the method based on new time-domain features yields higher correct classification of the instrument classification numbers and recognition rate than the traditional features. Then we analyzed the single-label classification, and multi-tone polyphonic music label classification problem is studied, using of the traditional multi-label classification and decision tree algorithm, a combination of these two types of algorithms algorithm, decision tree based on multi-label (ML-Decision Tree) in the classification algorithm, the experimental results show that the classification algorithm based on multi-label decision tree (ML-Decision Tree) yields higher recognition rate than the single-label classification algorithm. And using of the traditional multi-label classification and K nearest neighbor algorithm, a combination of these two types of algorithms algorithm, based on multi-label K nearest neighbor algorithm (ML-KNN) classification algorithm, the experimental results show that the classification algorithm based on multi-label K nearest neighbor algorithm (ML-KNN) yields higher recognition rate than based on multi-label decision tree (ML-Decision Tree).