Loading...
|
Please use this identifier to cite or link to this item:
https://nccur.lib.nccu.edu.tw/handle/140.119/158309
|
Title: | 人工智慧的樂器自動調音系統研究與實現 Design and Realization of an AI–Driven Automatic Instrument Tuning System |
Authors: | 劉宇榛 Liu, Yu-Chen |
Contributors: | 黃國峯 林日璇 Huang, Kuo-Feng Lin, Jih-Hsuan 劉宇榛 Liu, Yu-Chen |
Keywords: | 人工智慧 卷積神經網路 自動調音 馬林巴木琴 基頻偵測 Artificial Intelligence CNN Automatic Tuning Marimba F0 Estimation |
Date: | 2025 |
Issue Date: | 2025-08-04 12:59:06 (UTC+8) |
Abstract: | 本研究驗證人工智慧於馬林巴木琴音板自動調音之可行性。系統以卷積神經網路(CNN)執行基頻(F0)偵測,並整合時頻預處理與 PID 控制,建構「敲擊-量測-加工」的閉迴路原型。離線測試顯示 F0 誤差可收斂至 ±10 cent;然於即時環境中,非諧和泛音與瞬態峰值放大誤差至 ±15 – 25 cent,致使控制迴路失穩。雖未能完全達成全自動調音,試驗結果證實:均質玻璃纖維音板可降低模型變異,並揭示需在高魯棒 F0 網路、模型預測控制與數位雙生仿真等面向持續突破。研究所提供之量化基線與反思,對後續打擊樂器智慧製造具有參考價值。 This study investigates an AI-assisted automatic-tuning prototype for marimba bars. A convolutional neural network (CNN) estimates the fundamental frequency (F0); the output is fed to a PID-based controller that emulates material removal. With ≈2 000 labelled strikes, the CNN attains ±10 cent offline accuracy, yet degrades to ±15–25 cent in real-time due to inharmonic overtones and percussive transients. Control loops therefore fail to converge. Although full automation was not achieved, the work quantifies key obstacles and outlines future directions in robust F0 estimation and non-linear control, providing a data baseline for smart marimba manufacturing. |
Reference: | 1.Garcia, P., Li, H., & Wang, Y.(2024)。〈Objective timbre metrics for percussive instruments〉。Journal of the Acoustical Society of America, 156 (1), 45–57。(引用範圍:pp. 50–52,提出 T60 與頻譜質心量測公式) 2.Helmholtz, H.(1877)。On the Sensations of Tone(4th ed.)。New York:Dover。(引用範圍:pp. 123–130,音程與拍頻理論) 3.Hochreiter, S., & Schmidhuber, J.(1997)。〈Long short–term memory〉。Neural Computation, 9 (8), 1735–1780。https://doi.org/10.1162/neco.1997.9.8.1735(引用範圍:pp. 1739–1742,LSTM 結構說明) 4.Jones, J. M.(2019)。〈Manual tuning practices in professional marimba craftsmanship〉。Percussive Arts Journal, 57 (2), 34–42。(引用範圍:p. 36,4–6 年訓練時程數據) 5.Kim, H., Lee, J., & Nam, J.(2019)。〈CREPE: A convolutional representation for pitch estimation〉。收錄於 IEEE ICASSP 2019(pp. 161–165)。IEEE。https://doi.org/10.1109/ICASSP.2019.8683852(引用範圍:p. 163,RPA ≈ 92 % 結果) 6.LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P.(1998)。〈Gradient–based learning applied to document recognition〉。Proceedings of the IEEE, 86 (11), 2278–2324。https://doi.org/10.1109/5.726791(引用範圍:pp. 2281–2283,CNN 基本原理) 7.Ohashi, K., & Kato, S.(2022)。〈Decay characteristics of tropical hardwood xylophone bars〉。Applied Acoustics, 195, 108752。https://doi.org/10.1016/j.apacoust.2022.108752(引用範圍:Table 3,玫瑰木 T60 ≈ 3.5 s) 8.Rabiner, L., & Schafer, R.(1978)。Digital Processing of Speech Signals。Englewood Cliffs, NJ:Prentice–Hall。(引用範圍:pp. 368–372,倒頻譜基頻偵測) 9.Ross, D., Chen, M., & Zhou, Q.(2023)。〈Impact of striking position on marimba bar resonance〉。收錄於 Proceedings of ISMIR 2023(pp. 120–127)。(引用範圍:Fig. 2,擊點 22 % vs 50 % 比較) 10.Smith, A., & Chang, L.(2023)。〈Pitch variance tolerances in modern marimba manufacturing〉。Percussive Arts Journal, 61 (3), 12–18。(引用範圍:p. 14,±5 cent 容許值) 11.Taylor, T. D.(2002)。Strange Sounds: Music, Technology & Culture。New York:Routledge。(引用範圍:Chapter 4,樂器材料與文化影響) 12.Vincent, E., Gribonval, R., & Févotte, C.(2010)。〈Performance measurement in blind audio source separation〉。IEEE Transactions on Audio, Speech, and Language Processing, 14 (4), 1462–1469。https://doi.org/10.1109/TASL.2005.858005(引用範圍:pp. 1464–1466,SDR 評估指標) |
Description: | 碩士 國立政治大學 經營管理碩士學程(EMBA) 110932422 |
Source URI: | http://thesis.lib.nccu.edu.tw/record/#G0110932422 |
Data Type: | thesis |
Appears in Collections: | [經營管理碩士學程EMBA] 學位論文
|
Files in This Item:
File |
Description |
Size | Format | |
242201.pdf | | 1288Kb | Adobe PDF | 0 | View/Open |
|
All items in 政大典藏 are protected by copyright, with all rights reserved.
|