古琴琴音之合成與減字譜之解析-撥弦樂器物理模型分析與音樂合成的新方法

Abstract

過去幾年中，我們針對中國最古老的撥弦樂器─古琴，進行實體分析與聲音合成的相關研究。古琴具有超過1500種以上常用以及數千種必備但較不常用的彈奏方法，可視為世上彈法最複雜的撥弦樂器。本論文提出我們針對古琴所進行一連串研究所獲得的成果。首先進行振動機構的模擬。傳統的古琴琴弦由蠶絲纏繞而成，而近代的琴弦則以蠶絲或尼龍絲包覆的鋼弦為主。為了量測琴弦的振動特性，我們設計一個鋼弦量測系統以獲得相關動態資訊。我們亦提出一個以琴弦實體為基礎的類神經網路─Scattering Recurrent Network (SRN) 用以模擬與分析不同材質的琴弦比如古琴，琵琶，吉他等中西樂器所用琴弦的振動特性。接著我們進行古琴音樂的合成。由於特殊的彈奏技巧比如滑音與顫音的頻繁使用亦是古琴音樂的特色之一，因此為能夠產生逼真的古琴音樂，將滑音彈奏裝置融入實體合成模型為一有效的解決方案。我們提出 Modified Scattering Recurrent Network (MSRN) 作為撥弦樂器合成之通用模型。我們亦發展一個新的學習方法可加速模型的學習並獲得更佳的合成參數。針對此模型我們亦內建一有效率的滑音合成法於合成模型使之可輕易產生古琴音樂廣泛使用的滑音彈奏效果。最後設計一古琴電腦作曲系統。一電腦輸入介面可組成古琴所特有的漢字樂譜─減字譜，並以BMP檔案形式儲存。藉由琴譜的分析，MSRN合成模型可由查表方式獲得所需之合成參數並自動產生使用者所想要的古琴合成音樂。

In the past few years, we have focused on the analysis and synthesis of a plucked-string instrument called the Chin, one of the oldest Chinese instruments. There are around 1500 playing techniques that are frequently used for the Chin. There are still thousands more that are used in a less regular base. It is acknowledged to be the most complicated of all the string instruments of the world. In this dissertation, a series of researches is developed on this instrument. The first is the modeling of the vibration mechanism of stringed instruments. Traditional Chin strings are made of pure silk and modern Chin strings are either nylon or silk wound round solid steel strings. A steel-string measurement system is designed to measure the vibrations of musical strings. A model-based neural network called Scattering Recurrent Network (SRN) is proposed to model and analyze the dynamics of various acoustic strings such as Chin strings, guitar strings and Pipa strings. The second is the music synthesis research with respect to the Chin. In addition, special effects such as vibrato and wide range portamento are frequently used. It is necessary to incorporate these effects in the synthesis method in order to produce realistic synthesis Chin music. The Modified Scattering Recurrent Network (MSRN) is proposed as a general synthesis method for plucked-string instruments. A new training method is developed to speed up the training and obtain better system parameters. A MSRN embedded technique that can produce special playing techniques such as vibrato and portamento is also described. The final is the computer-aided composition system for the Chin. A computerized input interface is constructed to compose Chien-Tzu Pu (the Chin tablature) and the tablature can be stored in BMP format. In addition, by analyzing the Chin tablature, the MSRN synthesis system can look up the parameter table to obtain the synthesis parameters and automatically produce the required synthetic Chin music.