How to add Brazilian Portuguese?
The Portuguese in the model generates only Portugal Portuguese that is very different from how Brazilians speak, not only phonetically, but grammatically too.
Even when I upload a sample of my voice, it's me with a strong Portugal Portuguese accent, with suppression of syllables and fast speech tempo, like in Portugal.
It sounds very different from Brazil that we speak every syllable, following the cadence established by the diacritics "á, à, â, ã, é, ê, ó, ô,ã,ú, ç"... also, we don't have short "p" or "c", also in Brazil, normally we speak the "L" in the end of a world with the sound close to the sound of "U".
There are a lot of differences, these are ones that the listener can identify it's not from Brazil, but Portugal or other Lusophone country by just hearing 3 or 4 words.
I tried dozens of seeds, but none fit.. There's a specific seed or configuration I can set? Or there's a way to add.
Thank you all.
You can finetune the model under the guidance of the documents.
The Portuguese in the model generates only Portugal Portuguese that is very different from how Brazilians speak, not only phonetically, but grammatically too.
Even when I upload a sample of my voice, it's me with a strong Portugal Portuguese accent, with suppression of syllables and fast speech tempo, like in Portugal.
It sounds very different from Brazil that we speak every syllable, following the cadence established by the diacritics "á, à, â, ã, é, ê, ó, ô,ã,ú, ç"... also, we don't have short "p" or "c", also in Brazil, normally we speak the "L" in the end of a world with the sound close to the sound of "U".
There are a lot of differences, these are ones that the listener can identify it's not from Brazil, but Portugal or other Lusophone country by just hearing 3 or 4 words.I tried dozens of seeds, but none fit.. There's a specific seed or configuration I can set? Or there's a way to add.
Thank you all.
That's Interesting because the model used at https://fish.audio/app has very nice and correct pronunciations Brazilian voices (e.g. search for "Cid Moreira" public voices) to test. I think the S1 proprietary model used there is best than the distilled S1 avaliable to us.