請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/10094
標題: | 以結合不同分類法與類神經網路為基礎在吉他和弦即時辨識器之比較 Guitar chord real-time recognition system based on different classifiers and Neural Network |
作者: | Chih-Hung Wang 王智弘 |
指導教授: | 陳國在 |
關鍵字: | 類神經網路,貝氏分類法,Knn分類法,吉他和弦,和弦辨識, Neural Networks,Na&iuml,ve Bayes Classifier,K-th nearest neighbor classifier,guitar chord,chord recognition, |
出版年 : | 2011 |
學位: | 碩士 |
摘要: | 音樂是人們生活中不可或缺的文化元素,隨著科技進步,樂音辨識伴隨著不少實用的價值,諸如:卡拉ok音準評分機、電子調音器、自動翻譜機…等都是樂音辨識的應用。
本研究著重在樂音辨識中的和弦辨識一環,本論文採用木吉他和弦,以類神經演算法為核心,並搭配其餘兩種演算法,與個人電腦結合,讓使用者可以透過使用者介面,搭配視訊麥克風,即可進行吉他和弦即時辨識。論文首先建立好吉他和弦的資料庫,接著分析三種不同辨識分類法的辨識程度,並做比較與討論。 實驗分成兩種架構進行,第一種架構利用建立好的25000筆吉他和弦資料庫,共96種不同和弦,比對三種辨識分類法的辨識程度。實驗結果發現,類神經演算法在吉他和弦資料庫的平均辨識度可達99.26%,為三種分類器中辨識度最高者;貝氏分類法平均辨識度則可達94.59%,Knn分類法平均辨識度為75.46%。 第二種架構為分別測試吉他和弦即時進行下所辨識的程度,採用四種不同的和弦進行,並使用四種不同音色的樂器(兩種真實木吉他,兩種虛擬音源)進行辨識度比較。實驗結果發現,類神經演算法在四種樂器的平均辨識度可達80.58%,為三種分類器中辨識度最高者;貝氏分類法平均辨識度則可達68.86%,Knn分類法平均辨識度為54.01%。 In the culture for human life, music is necessary and elementary. Following the progress in science and technology, musical recognition, such as karaoke pitch-scoring machine, automatic tuning machine of music and electronic tuner, etc., did present a lot of practical values. In this study, it is to concentrate on the chord recognition, which is a part of the music-recognition technology. Thus, this study tries to use acoustic guitar chords to make in-real-time recognition by users when using user interface to match with microphone through the combination of three different recognition-classification algorithms with personal computer. Accordingly, this study at first sets up the database for the guitar chord, and further analyses the recognition percentage for three different recognition-classification algorithms as used and then makes comparison and discussion. Regarding to the experiment, two prototypes are used to perform. Between them, the first prototype uses the database of the guitar set up beforehand, which overall have ninety six items, to perform recognition percentage by using the above three different recognition -classification algorithms. The results as obtained for experiment find the averaged recognition percentages by 99.26, 94.59 and 75.46 of the guitar-chord database by using, respectively, neural network-, Naïve- Bayes classification- and Knn-algorithms are reached. As regards to the second prototype, it, respectively, makes in real time recognition percentage of guitar chord in progress by using four different chords created, respectively, by four musical instruments of different acoustic characteristics, which include two actual acoustic guitars and two virtual sound sources, and then compare the results as obtained for recognition percentages. From the result as obtained from the experiment, it can find the associated averaged recognition percentages by 80.58, 68.86 and 54.01 to the four musical instruments by, respectively, using neural network-, Naïve-Bayes classification- and Knn-algorithms are reached. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/10094 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 工程科學及海洋工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-100-1.pdf | 9.84 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。