Has anyone worked on fine-tuning models for different English accents?

[NehaB18]

Thanks for the awesome model.

Default accent looks like American English. Has anyone worked on fine-tuning models for different English accents? like Indian English or Irish English?

I am interested to know how much dataset and fine-tuning will be required to achieve this?

[lfhuhu]

Hello, I wonder if you have any experience in doing English with accent these days. Thank you very much

[piotrwitkowski]

Should be possible to fine-tune it with a good dataset.
I have some success, but probably would need more samples to make it work better.
FF5-TTS dwarwish/scottish

[piotrwitkowski]

As for the dataset size, I noticed that anything below 500 samples had issues with inference. That said, my fine-tuning with 1k samples performed better than the 8k one—I’d say it largely depends on dataset quality. Or maybe it’s just a skill issue on my end.