Font Size: a A A

Based On The Research Of Humming Music Retrieval System And The Implementation

Posted on:2013-03-31Degree:MasterType:Thesis
Country:ChinaCandidate:Y LiFull Text:PDF
GTID:2248330374985609Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Humming based music retrieval is a technology with broad application prospectsthat is talked a lot in recent years. The target of this thesis is to build a fully functionalhumming based music retrieval system.In the research work of the humming based music retrieval system, we mainlyfocus on two main aspects. The first aspect is the handling of the humming audio andthe melody extraction process. The second aspect is the research of melody matchingalgorithm. These two aspects are the most important part of the humming based musicretrieval system, they are also the main points of this thesis.In the research of the first aspect, we firstly did some pretreatment to the audioincluding the process of reducing noise, pre-emphasis, and separating the audio intoframes. We used the combination of two methods to reduce noise, one of which was toset a threshold to the volume and the other was the spectral subtraction method. Andthen we used a high-pass filter to pre-emphasis the audio. We used the hammingwindow to separate the audio into frames. We used existing methods to complete theprocess of the pretreatment of the audio. We did a lot of research work on the melodyextraction. We did a lot of simulation experiments, and researched kinds of existingmelody extraction algorithms. And then, we chose a classic ACF method and madesome improvements to it, and completed the process of melody extraction with it. Inthe same time, we did some research work on the melody extraction of MIDI formattedmusic to build a music characteristic database. We combined some advantages of somealgorithms to make a MIDI melody extraction model, and built the database files.For the second aspect, we did a lot of research on it. Firstly, we researchedmelody matching algorithms based on string matching, melody matching algorithmsbase on line scaling, and melody matching algorithms based on statistics separately.And at last, considering our needs, we chose a dynamic programming based melodymatching algorithm named DTW(Dynamic Time Warping) algorithm. We improvedthe algorithm based on a lot of simulation experiments. And we performed theimproved algorithm in our humming based music retrieval system. Finally, we built a fully functional humming based music retrieval system on theVS2008platform with c programming language. And then, we did a lot ofexperiments and data analysis. The system completed the humming based musicimformation retrieval function with satisfactory performance. It achieved ourexpectation.
Keywords/Search Tags:query by humming, melody extraction, melody matching, ACF method, DTW
PDF Full Text Request
Related items