How to deal with classic southern england accent

[victorrosi]

Dear Tortoise community and author,

Thank you for this brilliant tool! It is really easy to use for me, as I am a poor research in cognitive science that has no background in deep learning.

In a project, I aim to evaluate voice cloning technologies within a voice identity perception experiment, with voices displaying a classic southern England accent. However, it seems that with a few samples from british participants (while trying to maximize the phonetic diversity) the synthesised audio will always sound more American. I noted that is something mentioned by the author in the README: "... Tortoise will be particularly poor at generating the voices of minorities or of people who speak with strong accents.".

But as it is a constraint I cannot avoid in my work, I would like to now if there is a solution for me to obtain the adequate accent in synthesised audio if my original voice samples are from UK people. Also, if it is necessary for me to learn how to train the model on an other dataset, I will take all advice and support. Again, I know nothing about these tools, but I'm keen to learn).

Thanks for your help and your work!

Victor

[neonbjb]

Hey Victor, Thanks for the kind words.

I don't really have any good answers for you, unfortunately. This is just a limitation of this model. I do not intend to release any information on fine-tuning the model as I am concerned about the ethics of unleashing a near-perfect voice cloning program on the world.

[victorrosi]

Hi James, thank you for your answer :) !

The use of this tool would only be for Academic Research purposes. We will get ethical approval for any experiment using voice cloning anyway and everything will be in accordance to UK GDPR...

Is there a way to chat more about that with you? Victor