Turkish Vocabulary
Hi Everyone,
I have downloaded and used this Turkish tts model. Generally it is quite nice.
However, there are a few small problems.
From my point of view, the most important problem is that the punctuations seem to have no effect on the pronouncation.
Therefore, the synthesized speech intonation does not sound natural. For example comma, dot , question mark, exclamation mark all do not change the tonation. MMS just ignores all of them.
One possible reason that could be causing this unnatural behaviour is those punctuation marks are missing in the vocab.json file.
I think, since they are not in the vocab list, MMS does not learn them, so as a result, it ignores them.
Well, as a quick solution I added those punctuation chars into the vocab as below:
... existing character mappings
...
"Μ": 4, <<<< here the original list ends
I added the followings to the list:
",": 44,
"!": 45,
"?": 46,
";": 47,
":": 48,
"7": 49,
"8": 50,
"9": 51,
".": 52
So : I updated the vocab_size parameter in the config.json file as "vocab_size": 53, where its orignal value was 43.
the 43'th char is mapped to unk. So, I continued from 44.
When I run inference with this modified model, it throws unmatched error.
So, I could not actually make use of this modified vocab.
I tried to find a solution on the internet, but I could not find a solution for this problem.
But from my explorarions an idea came to my mind.
I modified the original model mms-tts-tur with the new vocab.
And I am going to re-finetune from this modified model, and see if I can make those punctuation characters have effect on the synthesized audios.
As soon as I try this, I will share the results here.
By the way, I just want to make sure that, I am on the right path.
If any of you has encountered this problem or a similar one, could you please share your experience and findings with me.
In summary what I want is to make my mms tts model as natural and humanly as possible.
Any recommendations on that will be much appreciated.
Thanks in advance.
Y.A.
Just as an additional side note:
In the original vocab, the digits from 0 to 6 are included, so, I wanted to complete it upto 9, that is why, I added 7,8 and 9.
other than that, 7,8, and 9 have no special purpose.
hi. can you inform us about the outcome please?
how did you modify the original model mms-tts-tur with the new vocab. did you modify the actual code or what