Could you replace the autoregressive.pth file with a custom file (still named autoregressive.pth) that was trained only on speakers with a specific accent (Southern, or German for example)... and then train an accented speaker against that? Or would it break.
The problem is the original custom file would have been trained against the original autoregressive file.
It seems then the processing doesn't "understand" the accent, it just makes it sound British. At best, accents are smoothed out, like an average between American English and whatever accent is applied...trying to get a thick accent on a few voices.
Could you replace the autoregressive.pth file with a custom file (still named autoregressive.pth) that was trained only on speakers with a specific accent (Southern, or German for example)... and then train an accented speaker against that? Or would it break.
The problem is the original custom file would have been trained against the original autoregressive file.
It seems then the processing doesn't "understand" the accent, it just makes it sound British. At best, accents are smoothed out, like an average between American English and whatever accent is applied...trying to get a thick accent on a few voices.