hexgrad

/

Kokoro-82M

like 2.83k

same here, the output isn't natural at all, unlike the Chinese demo provided before (was pulled down in the current demo)

Try replace full-width punctuations like ， 。 with half-width ones like , .

@hexgrad When preprocessing the text it may help to normalize all kinds of Unicode punctuations into a few ones. Besides comma and period, there are also issues like the French quotation marks « »

As a native Chinese, I think xiaobei and xiaoni sound like some regional dialects or foreigners speaking Chinese, and the other voices xiaoxiao, xiaoyi, yunjian, yunxi, yunxia, yunyang sound like standard Chinese

Try replace full-width punctuations like ， 。 with half-width ones like , .

@hexgrad When preprocessing the text it may help to normalize all kinds of Unicode punctuations into a few ones. Besides comma and period, there are also issues like the French quotation marks « »

As a native Chinese, I think xiaobei and xiaoni sound like some regional dialects or foreigners speaking Chinese, and the other voices xiaoxiao, xiaoyi, yunjian, yunxi, yunxia, yunyang sound like standard Chinese

Thanks a lot!

But how to work around from " the English words in the middle cannot be pronounced" ?

I guess that can be fixed by modifying the g2p pipeline

I guess that can be fixed by modifying the g2p pipeline

Thanks!