本文選題：樂(lè)器識(shí)別 + 特征抽取��； 參考：《華南理工大學(xué)》2012年碩士論文

【摘要】：近年來(lái)，隨著數(shù)字音樂(lè)創(chuàng)作、收集以及存儲(chǔ)技術(shù)的快速發(fā)展，許多機(jī)構(gòu)積累了大量的音樂(lè)音頻數(shù)據(jù)。如何對(duì)這些音頻資源進(jìn)行有效的組織和管理，使得人們能從大量音頻數(shù)據(jù)中進(jìn)行查詢和處理他們需要的音頻數(shù)據(jù)，已經(jīng)成為一個(gè)迫切的需求。在聲源識(shí)別中，語(yǔ)音信號(hào)處理與識(shí)別是一個(gè)傳統(tǒng)的研究熱點(diǎn)，隨著計(jì)算機(jī)識(shí)別技術(shù)的迅速發(fā)展，基于內(nèi)容的音樂(lè)信號(hào)分析也逐漸成為一個(gè)新的研究熱點(diǎn)。根據(jù)基于內(nèi)容的音樂(lè)特征研究，音色特征可以用來(lái)描述音頻文件音色，使基于內(nèi)容的音樂(lè)查詢搜索引擎的實(shí)現(xiàn)成為可能。而樂(lè)器識(shí)別是其中一個(gè)重要的應(yīng)用。樂(lè)器識(shí)別的應(yīng)用非常廣泛，可以應(yīng)用于基于內(nèi)容的音樂(lè)轉(zhuǎn)錄、音頻的結(jié)構(gòu)化編碼、音樂(lè)推薦及查詢引擎以及音樂(lè)評(píng)注等方面。 本論文詳細(xì)闡述了基于聲學(xué)特征的樂(lè)器識(shí)別的基本原理和實(shí)現(xiàn)過(guò)程。首先對(duì)樂(lè)器識(shí)別中常用的聲學(xué)特征做了較為深入的研究，并且詳細(xì)闡述了特征的提取方法。本論文實(shí)驗(yàn)使用了Mel倒譜系數(shù)（MFCC）、Mel差分倒譜系數(shù)(ΔMFCC)等作為樂(lè)器識(shí)別系統(tǒng)的特征系數(shù)，重點(diǎn)研究了支持向量機(jī)（SVM）的分類原理，并作為樂(lè)器識(shí)別的分類器算法。 在實(shí)驗(yàn)測(cè)試時(shí)，首先使用了不同的聲學(xué)特征系數(shù)對(duì)樂(lè)器進(jìn)行識(shí)別，從實(shí)驗(yàn)結(jié)果分析這些特征對(duì)樂(lè)器識(shí)別的正確率的影響。然后對(duì)實(shí)驗(yàn)進(jìn)行了進(jìn)一步的研究，使用了主成分分析技術(shù)（PCA）和陳森平等提出的改進(jìn)的最大間隔的支持向量機(jī)特征選擇算法，實(shí)驗(yàn)表明， PCA可以有效實(shí)現(xiàn)支持向量機(jī)分類器特征向量的簡(jiǎn)約，，縮短了樂(lè)器識(shí)別系統(tǒng)的訓(xùn)練和識(shí)別時(shí)間。通過(guò)實(shí)現(xiàn)的改進(jìn)的特征選取算法，則可以選取出其中最有效的特征，從而提高支持向量機(jī)的泛化能力，進(jìn)一步提高樂(lè)器識(shí)別的識(shí)別正確率。
[Abstract]:In recent years, with the rapid development of digital music creation, collection and storage technology, many organizations have accumulated a large amount of music audio data. How to organize and manage these audio resources effectively, so that people can query and deal with the audio data they need from a large number of audio data, has become an urgent need. Speech signal processing and recognition is a traditional research hotspot in acoustic source recognition. With the rapid development of computer recognition technology, content-based music signal analysis has gradually become a new research hotspot. Based on the research of content-based music features, timbre features can be used to describe the timbre of audio files, which makes the implementation of content-based music query search engines possible. Musical instrument recognition is one of the important applications. Musical instrument recognition is widely used in content-based music transcription, audio structured coding, music recommendation and query engine, music commentary and so on. In this paper, the basic principle and realization process of musical instrument recognition based on acoustic features are described in detail. Firstly, the acoustic features commonly used in musical instrument recognition are studied in depth, and the extraction methods of the features are described in detail. In this paper, the Mel cepstrum coefficients and the differential Cepstrum coefficients (螖 MFCC) are used as the characteristic coefficients of the musical instrument recognition system. The classification principle of support Vector Machine (SVM) is studied, and the classification algorithm is used as the classifier for the recognition of musical instruments. In the experiment, different acoustic characteristic coefficients are used to identify the musical instrument, and the influence of these characteristics on the accuracy of the instrument recognition is analyzed from the experimental results. Then the experiment is further studied, using the principal component analysis (PCA) technique and Chen Sen's equal improved support vector machine feature selection algorithm with maximum interval. Experiments show that PCA can effectively reduce the eigenvector of SVM classifier and shorten the training and recognition time of the instrument recognition system. Through the improved feature selection algorithm, the most effective feature can be selected, thus the generalization ability of SVM can be improved, and the recognition accuracy of musical instrument recognition can be further improved.
【學(xué)位授予單位】：華南理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2012
【分類號(hào)】：TP391.41

【參考文獻(xiàn)】

中國(guó)期刊全文數(shù)據(jù)庫(kù) 前9條

1 丁愛(ài)明;;作為說(shuō)話人識(shí)別特征參量的M FCC的提取過(guò)程[J];電子工程師;2006年01期

2 李翰;;中外樂(lè)器分類的研究[J];福建論壇(人文社會(huì)科學(xué)版);2008年S2期

3 張奇,蘇鴻根;基于支持向量機(jī)的樂(lè)器識(shí)別方法[J];計(jì)算機(jī)工程與應(yīng)用;2004年18期

4 陳啟買;陳森平;;支持向量機(jī)的一種特征選取算法[J];計(jì)算機(jī)工程與應(yīng)用;2009年23期

5 張一彬;周杰;邊肇祺;郭軍;;基于內(nèi)容的音頻與音樂(lè)分析綜述[J];計(jì)算機(jī)學(xué)報(bào);2007年05期

6 鄭怡文;;典型的音頻分類算法[J];計(jì)算機(jī)與現(xiàn)代化;2007年08期

7 曾黃麟,虞厥邦,曾謙;基于主成分分析的特征簡(jiǎn)化[J];四川輕化工學(xué)院學(xué)報(bào);1999年01期

8 但志平;胡剛;劉勇;;基于LPC倒譜參數(shù)分析的說(shuō)話人識(shí)別系統(tǒng)[J];三峽大學(xué)學(xué)報(bào)(自然科學(xué)版);2007年01期

9 劉雅琴;智愛(ài)娟;;幾種語(yǔ)音識(shí)別特征參數(shù)的研究[J];計(jì)算機(jī)技術(shù)與發(fā)展;2009年12期

中國(guó)碩士學(xué)位論文全文數(shù)據(jù)庫(kù) 前1條

1 白亮;音頻分類與分割技術(shù)研究[D];國(guó)防科學(xué)技術(shù)大學(xué);2004年

本文編號(hào)：1816138