本篇論文主題為建構一個使用MPEG-7音訊特徵值(Audio Descriptor)作為搜尋依據的歌曲檢索系統,因為MPEG-7音訊特徵值取樣完具有大量的資料點,為了達到快速搜尋的原則,本論文將討論使用獨立成份分析(Independent Component Analysis)和因素分析(Factor Analysis)等來探討對於降低資料量維度的效果,以大幅降低搜尋時間且保有良好的辨識率;另外我們也對於系統做實際錄音測試,以及在各種不同的訊雜比下其辨識成功率,在有雜訊的狀況下該如何提高其辨識成功機率,以及判斷當歌曲不在資料庫中是否拒絕機制可以準確判斷;最後系統將錄音資料使用多重解析度搜尋技巧增加搜尋時間及準確性。
The thesis studies the use of MPEG-7 audio descriptor as features for music retrieval system. Since the MPEG-7 audio descriptors representing a song contains a large number of data points, this thesis uses ICA (Independent Component Analysis) and FA(Factor Analysis) for dimension reduction to reduce the search time and to maintain a good recognition rate. In addition, we also test the indentification capabilities of the audio descriptors with speaker-mic recorded music and music with artifical noise. To be a retrieval system, we also propose a method to determine whether a piece of music is in the database or not.