- 作者: 古鴻炎; 許文龍
- 作者服務機構: 國立台灣科技大學電機工程技術系
- 中文摘要: 在許多文句翻國語語音的系統裡,都採用音節為語音合成之單位,因此本文針對國語音節信號合成的問題提出了一個新的合成方法,與其它時域合成方法比較,除了能夠合成出清晰的語音信號之外,還提供了較多的信號控制之彈性,包括音長之控制,以調整說話速度及反映其它因素對音節長度之影響;音調(或基週軌跡)之控制,以便由第一聲音節去合成其它聲調的音節,而能夠節省記憶需求;以及聲道長度之控制,以便使男生原音合成出的女生聲音(或卡通人物的聲音)較為自然。我們發現當適當地調整聲道長度與音調高低,就可合成出許多互不相同的音色,聲道長度是一項新提出的控制因素,它使得音色的控制成為實際可行。雖然其它時域合成方法也有提供音調、音長之控制,但是我們的合成方法提供的彈性較高·且已讓這兩個控制因素產生的對共振峰軌跡的干擾降低很多。
- 英文摘要: The syllable is commonly adopted as a synthesis unit in Mandarin text-to-speech systems. Therefore,the problem of syllable signal synthesis is studied and a new synthesis method is proposed in this paper.When compared with other time-domain synthesis methods, our method not only can synthesize clear speechsignals,but also provides more flexibility in control of the duration, tone (pitch contour),and timbre.The duration of a syllable should be adjustable to reflect the influence of pronunciation speed and otherrelevant factors. The tone should be changeable because it is desired that syllables of other tones canbe synthesized from the First-tone syllable to save memory. The vocal track length intrinsic in a syllable'swaveform should be adjustable because it is desired that the speech of females (or cartoon actors) synthesizedfrom the original syllable signals of a male be perceived as more natural. We find that many distincttimbres could be synthesized if both the vocal track length and pitch-contour's height were adjustedsimultaneously. Here, the vocal track length is a newly studied factor and timbre control becomes realizableusing this factor. Although the ability to control a syllable’s duration and pitch contour is also providedby other time-domain synthesis methods, our method provides more flexibility and largely decreases theinterference the traces of formant frequencies incurred due to these two factors.
- 中文關鍵字: speech synthesis; Mandarin speech; waveform interpolation
- 英文關鍵字: --