在這篇論文中，提出了一個利用非負矩陣分解法(NMF)去定量分析多音音樂的方法。音樂訊號首先被轉換成頻譜圖(short time magnitude spectrum)，然後會根據樂譜的資訊，利用音樂色譜(chromagram)來做初步的比對工作，其目的是把音檔分割成較小的片段，以便獲得較準確的分析結果。之後，NMF會被套用到這些已分割的音樂片段，以便取得這個錄音檔中的每個音符的頻率和能量的相關資訊。最後，我們會得到每個音符的起始時間、結束時間和音色等等資訊。在我們提出的這個方法中，根據經驗所做的人工校正仍然被採行。目前而言，主要是針對鋼琴樂器。我們打算去產生一個可以信賴的資料庫，用以測試分析不同的音樂訊號分析演算法。

In this thesis, a quantitative analysis method for polyphonic musical recordings using Non-negative Matrix Factorization (NMF) is presented. Signal is first converted to short time magnitude spectrum. Then, preliminary alignment with respect to score information using chromagram is applied in order to separate the recording into smaller segments to have more accurate analysis. Finally, NMF is applied to each segment to have both spectral and intensity information for each note played in the recording. Finally, onset, offset and 11timbre information are generated. In the proposed method, manual correction is applied after examination of human experts. At the current stage, we focus on piano instrument. We intend to generate a reliable database for testing music signal analysis algorithms.

[1] "MIREX," http://www.music-ir.org/mirex/wiki/MIREX_HOME.
[2] Valentin Emiya (Telecom ParisTech. "MAPS Database – A piano database for multipitch estimation and automatic transcription of music," http://www.tsi.telecom-paristech.fr/aao/en/2010/07/08/maps-database-a-piano-database-for-multipitch-estimation-and-automatic-transcription-of-music/.
[3] P.-Y. Tsai, “Note-based Alignment Using Score-driven Non-negative Matrix Factorization,” The Department of Computer Science and Information Engineering (CSIE), Nation Cheng-Kung University, 2011.
[4] N. Hu, R. B. Dannenberg, and G. Tzanetakis, “Polyphonic Audio Matching and Alignment for Music Retrieval,” in in Applications of Signal Processing to Audio and Acoustics, IEEE Workshop., 2003, pp. 4.
[5] B. Niedermayer, and G. Widmer, "A multi-pass algorithm for accurate audio-to-score alignment."
[6] N. Orio, and D. Schwarz, "Alignment of monophonic and polyphonic music to a score." pp. 155-158.
[7] F. Soulez, X. Rodet, and D. Schwarz, "Improving polyphonic and poly-instrumental music to score alignment." pp. 143-148.
[8] L. Rabiner, and B. H. Juang, Fundamentals of speech recognition: Prentice-Hall, Inc., 1993.
[9] E. J. Keogh, and M. J. Pazzani, “Derivative dynamic time warping,” in In First SIAM International Conference on Data Mining, Chicago, Illinois., 2001.
[10] M. A. Bartsch, and G. H. Wakefield, “Audio thumbnailing of popular music using chroma-based representations,” IEEE Transactions on Multimedia, vol. 7, pp. 9, 2005.
[11] S. Salvador, and P. Chan, “Toward accurate dynamic time warping in linear time and space,” Intelligent Data Analysis, vol. 11, pp. 561-580, 2007.
[12] D. D. Lee, and H. S. Seung, “Learning the parts of objects by non-negative matrix factorization,” Nature, vol. 401, pp. 4, 1999.
[13] D. D. Lee, and H. S. Seung, “Algorithms for non-negative matrix factorization,” Advances in neural information processing systems, vol. 13, 2001.
[14] E. Battenberg, and D. Wessel, “Accelerating nonnegative matrix factorization for audio source separation on multi-core and many-core architectures,” in in 10th International Society for Music Information Retrieval Conference, Kobe, Japan, 2009.
[15] P. Smaragdis, and J. C. Brown, “Non-negative matrix factorization for polyphonic music transcription,” in in Applications of Signal Processing to Audio and Acoustics, 2003 IEEE Workshop on., 2003, pp. 177-180.
[16] Y.-L. Chen, T.-M. Wang, W.-H. Liao et al., “Analysis and Trans-synthesis of Acoustic Bowed-String Instrument Recordings: a Case Study using Bach Cello Suites,” in Dafx 2011, 9.10~9.23, Paris, 2011.