歌聲是造物者給於人類的一個祝福,人類可以透過歌聲來抒發其情感,好比快樂、憤怒、悲傷等等。然而,到目前為止,合成音色訊號的研究卻不多見。因此,這篇論文將試著合成音色訊號,並且使大部分的工作由電腦自動完成。本文將複雜的音色訊號合成分成兩部分去說明,分別是分析以及合成。分析部分:從被分析的訊號中萃取出數種參數,分別是樂譜參數、特徵參數以及聲音參數。在取得足夠的分析資料後,再依照文中的三種合成方法,利用這些參數去造出最後的合成歌聲。本文取材自歐婉菁學姊部分碩士論文,作者將其論文中看懂的部分整理記載成文。程式來源為計劃NSC 93-2213-E-002-081,後經研究助理彭智楹學長略加修改,交予作者使用。使用心得記載於文中。其中Wigner-Ville Distribution是作者加進去的部分,此部份已有初步的成果。
Singing voice is a blessing to humankind gifted by the Creator. Humankind can express their emotions, such as happiness, sorrow, anger, and etc., by singing a song. Unfortunately, to our knowledge, pertinent reasearches on this competency such as synthesizing voice signal are not common. Here, we investigated the nature of timbre and tried to computerize the process of human singing. This thesis is a summary of our studies on how to analyze and synthesize voice signal. We will elaborate on explaining how the human voice is parameterized by notation, characteristic, and voice parameters and how an intact voice can be retrived via these parameters. This thesis refers to the master thesis of Wan-Chin Ou. The program is from NSC 93-2213-E-002-081, and modified by Zhi-Ying Peng. The new contribution of this work is the Wigner-Ville Distribution upon which we have some preliminary results.