The Study Of Feature Extraction On The Query By Humming System

With the growing demand for the multimedia information, the corresponding multimedia retrieval technology has also become a research hotspot. The multimedia audio as an important resource, its search technology has become a key topic of concern related to the researchers. Music as a way for emotional expression and delivery, the research on it will be much more close to people’s living. For example, we may encounter such a scene in life, overheard or hum a song, which you are very familiar with, but can’t remember the song name, lyrics and complete information anyway, query by humming retrieval technology can solve this problem, by sing a melody of a song, the QBH system can return a ranked list of the songs which is more similar to the query song.Query By Humming retrieval is a kind of contend based music retrieval, it contains two modules that is feature extraction and melody match. Feature extraction module in the QBH has a very important position, whether the feature selection is highly independent, distinguished and representative or not has a far-reaching impact on the system search results. The paper describes the melody extraction of MIDI music database and humming feature including pitch and note extraction and processing, we make an assessment of the feature extraction by experiment. At last we obtain the whole system retrieval precision from the overall perspective of the QBH system.