The demos were generated by the following four models.
- FS2: the FastSpeech2 baseline model
- K0: the proposed method without discourse-level modeling (i.e. K=0)
- K10: the proposed method considering 21 neighbouring sentences in discourse-level modeling (i.e. K=10)
- K10 w/o BERT: the proposed method considering 21 neighbouring sentences in discourse-level modeling but without using BERT embeddings
Due to copyright issues, we cannot show the ground-true wavs of our data.
至今想起,那实在是青云一脉最悲苦的一段日子。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
其后,青叶正式出家,以本家姓叶,取青云之青字,故名青叶。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
整整一个晚上,在初见面那一刻的诧异与隐隐的激动过后,他们二人之间,便是长久的沉默。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
那声音凄厉而凶悍,声动四野,瞬间所有的黎族战士一起嘶吼,纷纷涌上。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
而是在南疆秘密流传了数千年的,那个神秘恶魔的传说。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
夺回镶在苗人圣器黑杖上的骨玉祭祀伟大的熊神。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
却是他不小心间又误入那片空地之中,只是此刻,却再也没有那股暖气起而抵挡了。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
他全身皆抖,手足无力,脚下一软,已瘫坐于地。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
片刻后一声巨响,两件大凶煞之物撞到一起,张小凡如受巨震,整个人被向上震起了一丈多高。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|
只是这雾气却又和内泽边缘上的瘴气不同,颜色并非灰色,已然是纯白模样。
FS2 |
K0 |
K10 |
K10 w/o BERT |
|
|
|
|