本論文的研究動機是要讓鋼琴初學者在使用手機上節拍器軟體的時，同時以手機錄製其演奏練習，在演奏結束後對此演奏加以評分，讓初學者能知道自己練習的優劣。本研究目的就是在優化此評分系統，讓手機系統能夠正確的評斷鋼琴的分數。
　　此評分系統設計是讓手機能對鋼琴演奏的錄製作評分，所以本論文中所使用的音檔資料皆為手機所錄製，在錄製的同時手機也撥放著節拍器聲，音檔的長度皆為14秒，且因所設計的評分流程所需，前四拍的節拍聲為空拍，不得彈奏鋼琴。
　　本論文主要的方向是先設計出一套評分流程流程，此實作流程分四步驟：（1）利用傅立葉轉換擷取乾淨節拍聲片段的頻率能量分布、（2）設立門檻挑選欲刪除之頻段 (3)使用起點偵測擷取鋼琴聲的時間點 (4) 計算鋼琴聲與節拍聲的時間差並轉換為分數。在實驗結果與分析中，實驗一會針對此初步流程作資料評分；實驗二對刪除頻段的擷取作錯誤分析，也就是是調整第二步驟的部分；實驗三及四對節拍聲沒砍乾淨的殘留值作調整及錯誤分析，其調整的部份主要位於第三步驟的階段；實驗五最後將各個參數作最佳化，以求最準確的評分系統。
　　從以上之步驟及實驗中可以很明顯地看到分數高低的判斷是依據節拍器正拍上是否有鋼琴聲為依據，因此許多流行歌曲中，鋼琴聲不是位於正拍上，更甚至位於反拍上，此種鋼琴演奏不會列入此實驗中的考量。
最後實驗結果顯示，在經過多次的錯誤分析及參數調整後，趨勢誤差值已從實驗一的34.66降至降至實驗五的20.71，且資料分佈趨勢合理，可以說此評分系統對於初學者的鋼琴演奏練習是有幫助的。

　The researches have revealed how beginning pianists can　analyze their performance quality by using the software　which is called metronome , players record their melodies　with the cell phone software. At the end of piano playing , the software will show a score that tell pianist weather　this piano playing is terrific or hideous. The main idea of our researches is to make the grading system in this　software as correctly as possible. As well as recording the performance, players can hear the beat transmit from cell phone. There are 14 seconds only in each audio data, the　former quarter of tempo should be blank without piano　sounds.
　There are four steps to accomplish the grading system: the first one is by utilizing Fast Fourier Transform to capture the power distribution of spectrum gathering from clear　metronome tick. Secondly is to establish the threshold　choosing the frequency bin, whose energy exceed the　threshold, tempted to eliminate. Third one is to use onset detection sensing the moment when piano button is pressed. The last one is to calculate the timing difference between metronome tick form cell phone and piano onset ,　furthermore, translate it to points.
　As a consequence, our first experiment may show the audio data score distribution from our grading system. Second experiment will do the error analysis toward deleting frequency bin. Experiment three and four try to accommodate amplitude of the deleted tick as well as do the error analysis. Our final experiment will optimize all the parameters to make accurate grading system possible.
　After numerous times of error analysis and parameters regulation, average error has decreased from 34.66 to 20.71 since the first experiment. Apart from that, trend of data distribution is reasonable, we are able to confirm the grading system is beneficial to those pianists who are beginners.

[1] Daniel P.W.Ellis, “Beat Tracking by Dynamic Programming”, LabROSA, Columbia University, New York, 2007.
[2] James S. Walker, Faster Fourier Transform book (second edition), 1996.
[3] Juan Pablo Bello, Laurent Daudet, Samer Abdallah, Chris Duxbury, Mike Davies, and Mark B. Sandler, “A Tutorial on Onset Detection in Music Signals”, Senior Member, IEEE, 2005.
[4] M.F. McKinney, D. Moelants, M.E.P. Davies and A. Klapuri, “Evaluation of audio beat tracking and music tempo extraction algorithms, ” Journal of New Music Research, 2007.
[5] Fu-Hai Frank Wu, Tsung-Chi Lee, Jyh-Shing Jang, Kaichun K. Chang, Chun Hung Lu, Wen Nan Wang, “A two -fold dynamic programming approach to beat tracking for audio music with time-varying tempo”, ISMIR, 2011.
[6] R. Bellman, Dynamic Programing. Princeton University Press, 1957.
[7] Herman J. Blinchikoff, Anatol I. Zverev, Filtering in the Time and Frequency Domains book, 2001.