aihpi

/

F5-TTS-German

like 25

@Boutler I can confirm, the latest F5-TTS uses a different base-model, so using this one just generates gibberish. Would also love to see a newer version of the German model!

@Boutler Well, seems I just used the wrong cfg-options lol
Try it again with the old cfg-options: {"dim": 1024, "depth": 22, "heads": 16, "ff_mult": 2, "text_dim": 512, "text_mask_padding": false, "conv_layers": 4, "pe_attn_head": 1}

That fixed it for me, using the WebUI.

Downgrade F5-TTS to tag 0.6.2 will make this model work again.

@Boutler Well, seems I just used the wrong cfg-options lol
Try it again with the old cfg-options: {"dim": 1024, "depth": 22, "heads": 16, "ff_mult": 2, "text_dim": 512, "text_mask_padding": false, "conv_layers": 4, "pe_attn_head": 1}

That fixed it for me, using the WebUI.

@Dezenz
Sorry if I sound a bit silly.
Which WebUI do you mean? I used that model in Pinokio. I've read about the configuration before, but I don't know how or where to replace it...

/--------------------------------------------------------------------------------------------------------------/

Downgrade F5-TTS to tag 0.6.2 will make this model work again.

@ol-k
As described above, I use Pinokio. And I don't know how to download an older version there. I've tried it outside of Pinokio, but I can't get it to work... /:

I don't use pinokio.
I use it standalone ...
Just checkout the repo with git and the tag 0.6.2

@Boutler did the suggestions (using other cfg-optiones or using the standalone version) help you to solve the issue? If not, please provide more information about how you use F5-TTS and your error message.

@Boutler Ah okay, the Pinokio version doesn't seem to have a button for custom models, if I've researched this correctly. Maybe it's a rather old version of F5-TTS, I guess? If so, I'm afraid I can't tell you how to implement it... :(

Fortunately, installing the version from Github is quite easy, even if the documentation is not quite so friendly, honestly. But if you feed the Github website to ChatGPT, Gemini and co, you should be given a clear guide on how to install it. You just need to install conda before (or alternatively do it via venv with Python command). Then, it should work pretty much after cloning the repo and install the requirements with "pip install -e".