基於離散餘弦轉換之音高偵測系統

Abstract

本研究的目的是開發一個系統，幫助演奏家練習單音曲，同時減輕音樂老師的負擔。 我們提出了一個創新的解決方案，讓學生在不依賴音樂老師即時指導的情況下，隨時隨地進行練習。學生只需錄製自己的演奏並上傳到系統中，系統將自動分析並提供即時的音高評估和回饋。這樣一來，學生可以更靈活地安排練習時間，並在需要時獲得準確的指導。 在方法方面，我們使用了離散餘弦轉換和峰值檢測演算法。DCT用於將錄音數據從時域轉換為頻域表示，並且峰值檢測算法用於檢測頻譜中的主要音高。 研究結果顯示，系統具有出色的音高準確性和一致性。據我們的結果，以範例的Mazas No. 5練習曲與MIDI計算相異，系統的平均音高誤差小，並具有低標準差。 這個系統的開發不僅對學生有益，同時也能減輕音樂老師的負擔。音樂老師可以專注於更有價值的指導和支持方面，提供更專業和個性化的指導。

The purpose of this study is to develop a system (ViolinPitch) that helps musicians practice solo pieces while reducing the burden on music teachers, allowing them to focus on providing more accurate and valuable guidance. We propose an innovative solution that enables students to practice anytime, anywhere without relying on immediate guidance from a music teacher. Students simply record their performance and upload it to the system, which will automatically analyze and provide real-time pitch assessment and feedback. This allows students to have more flexibility in arranging practice time and receive accurate guidance when needed. In terms of methodology, we utilize the Discrete Cosine Transform (DCT) and the peak detection algorithm. The DCT is used to transform the recorded audio data from the time domain to the frequency domain representation, while the peak detection algorithm is employed to identify the main pitches in the spectrum. Based on our findings, the system exhibits a small average pitch error for the Mazas No. 5 etude compared to MIDI calculations, with a low standard deviation. The development of this system not only benefits students but also alleviates the burden on music teachers. It enables music teachers to focus on providing more valuable guidance and support, delivering more professional and personalized instruction, and better meeting the needs of their students.