I use fastpitch training using MFA toolkit, however with using MFA, whitespace characters are not generated duration, so when train or inference we are discarding all the space character in a sentence.
For example: “helloworld”
However, with Vietnamese, pairs of words with different pronunciations appear, but when removing the spaces, the same sentences appear.
For example, “khác ạ” and “khá cạ” are both pairs of words in Vietnamese, but when removing spaces, they will form “khácạ”. Therefore, it is confused when inference “khác ạ” to “khá cạ”.
Is there any way to solve this problem i am having with my language.