论文部分内容阅读
就目前语音合成自然度的现状,探讨了合成语音中韵律词边界V#C、VN#C之间的无声间隙和过渡音存在的问题,以及由此造成的合成语音中词或短语之间的顿挫感和个别音段自然度较差的问题。该文在基于对普通话协同发音生理(EPG)研究的基础上,揭示了韵律词边界存在的协同发音现象并提出了解决合成自然度问题的方案。结果表明:韵律词边界闭塞(GAP)和停顿(SP)的区别在于,停顿表现在元音韵尾无过渡音且时长延长,辅音无声段时间较长,而闭塞则不同;语料库中增加擦音前韵尾的标注信息作为合成的匹配规则,可以消除合成中擦音前的顿挫感;韵尾过渡音中舌前辅音前面的韵尾F 2上升,舌前辅音中的翘舌音/zh,ch,sh,r,l/使韵尾的F 3下降。舌根音、唇音和唇齿音使前面的韵尾F 2下降;语调短语的韵律词边界没有V#C、VN#C的过渡音且边界间是停顿而非闭塞,不存在协同发音现象。
In this paper, we discuss the existing problems of silent gaps and transitional tones between the vowel boundaries V # C and VN # C in synthesized speech, and the consequent problems between words and phrases in synthesized speech Frustration and individual sections of the natural degree of poor problem. Based on the study of Mandarin Phoneticization (EPG), this paper reveals the phenomenon of synergistic pronunciation existing at the boundary of prosodic words and proposes a solution to the problem of synthetic naturalness. The results show that the difference between the prosodic word boundary occlusion (GAP) and pause (SP) is that the pause is characterized by no transitional tone at the end of the vowel and longer duration of the vowel, longer consonants without silence, and different occlusions; The ending information of the end of the consonant is used as a matching rule to eliminate the frustration before the fricative in the synthesis. The end of the preceding consonant in the end of the transitional tail increases in F2, and the accent in the anterior consonant is / zh, ch, sh, r, l / make the end of the F 3 decline. The base phonetic, lip, and lip croaks decrease the front end F 2; the prosodic word boundaries of the phonetic phrases do not have the transitional tones of V # C and VN # C, and the boundaries are pause instead of occlusion. There is no co-articulation phenomenon.