11752 Project

Chinese Talking Clock

(cross language limited domain speech synthesis)


The clock grammar: informal, as in normal conversation.


Pinyin-to-Sound Rule: Each unit in pinyin is treated as a Chinese phone, plus these allophones:

a --> ae /i_n
o --> u /{a_}
e --> ae /{i, u:}_
i --> ri /{zh, ch, sh, r}_
i --> si /{z,c,s}_
u --> u: /y_

Use English phones to approximate Chinese phones:

Here's the English voice approximation of the sentence: (click on the sentence to hear the wav file)

xian4zai4 ling2chen2 ling2dian3 zhong1 bu2dao4

Tones are added to the lexicon as stresses.
The true Chinese recordings are then aligned with the prompts of English approximations to build the limited domain voice.
Units are selected among those with the same tone.


And samples of the final synthesized Chinese Talking Clock (pre-synthesized wav files):

xian4zai4 ling2chen2 san1dian3 zhong1 bu2dao4 (the time is now almost 3 in the morning)
xian4zai4 zhong1wu3 shi2er4dian3 yi2ke4 bu2dao4 (the time is now almost quater after 12)
xian4zai4 wan3shang4 jiu3dian3 zhong1 guo4yi5dian3 (the time is now a little after 9 in the evening)
xian4zai4 gang1hao3 xia4wu3 si4dian3 wu3shi2 (the time is now exactly four fifty in the afternoon)
xian4zai4 bang4wan3 wu3dian3 si4shi2 guo4yi5dian3 (the time is now a little after five forty in the afternoon)
xian4zai4 ling2chen2 ling2dian3 zhong1 bu2dao4 (the time is now almost midnight)



Xiaojin Zhu, 2000/5/15