Unable to use fine tuned models

[mortenscheel]

Version 2.0.0, using the WebUI via athomasson2/ebook2audiobookxtts:huggingface.

• Selecting one of the models from the "Fine tuned models" dropdown makes the conversion hang forever. The container's output is Loading TTS xtts model from {modelName}... but there's no network or CPU activity.
• Downloading the required files from https://huggingface.co/drewThomasson/fineTunedTTSModels/tree/main/xtts-v2/eng, zipping them and uploading them as Custom XTTS model, doesn't work either. After uploading, the progress indicator says Error: cannot unpack non-iterable async_generator object.

[ROBERT-MCDOWELL]

where did you find the code where there is"athomasson2/ebook2audiobookxtts:huggingface" ?
about custom upload, it's a bug that will be fixed in the next update (today or tomorrow)

[DrewThomasson]

Weird that docker image should be doing it for you.....

I'll see if i can update the docker image or something so it's actually based on the compose

And not the huggingface

[mortenscheel]

@ROBERT-MCDOWELL

ebook2audiobook/README.md

Lines 231 to 242 in 5abebec

	#### Running the Docker Container
	To run the Docker container and start the Gradio interface, use the following command:
	-Run with CPU only
	```powershell
	docker run -it --rm -p 7860:7860 --platform=linux/amd64 athomasson2/ebook2audiobookxtts:huggingface python app.py
	```
	-Run with GPU Speedup (Nvida graphics cards only)
	```powershell
	docker run -it --rm --gpus all -p 7860:7860 --platform=linux/amd64 athomasson2/ebook2audiobookxtts:huggingface python app.py
	```

[mortenscheel]

I don't know if it's related, but the container also fails to download the Danish model:

Loading TTS tts_models/dan/fairseq/vits model from tts_models/dan/fairseq/vits...
                     Bad tar file - https://dl.fbaipublicfiles.com/mms/tts/dan.tar.gz

That URL returns a 403 response with this body

<?xml version="1.0" encoding="UTF-8"?>
<Error><Code>AccessDenied</Code><Message>Access Denied</Message><RequestId>1Q2Q331NEY8XCFC0</RequestId><HostId>LUI3E4L2GuuI2uy6ORl2eVMwjA35DxEWCu0q4iRD/p6t/LNOnQn3/5mmMye5Oqw7TcywnNFq5W50ipuI0EuTAQ==</HostId></Error>

[ROBERT-MCDOWELL]

@mortenscheel I'm afraid I made the same mistake than Croatian language, it was supported only by a token library called spacy which can work with Danish language, but xtts-v2 and fairseq standard models not.
as Croatian we must search for a TTS engine compatible with Danish. we planned to implement all best TTS but don't know when it will be released. probably for the next version. I you know a TTS engine capable to manage Danish so let us know!

[ROBERT-MCDOWELL]

@DrewThomasson I just realized we forgot to update the pip package list on dockers file and missed to add regex package on requirements.txt and pyproject.toml. Pushed in the next PR.
with athomasson2/ebook2audiobookxtts:huggingface I always thing that ebook2audiobookxtts means the old version of eb2ab, maybe it would be nice to change the name to ebook2audiobook

[DrewThomasson]

Your right I'll see if I can change it without having to create a whole new docker hub thing

[DrewThomasson]

@ROBERT-MCDOWELL

Kk changed it

It now will grab from the new correctly named docker hub repo

[ThePagi]

FYI the Docker instructions in the readme don't work for me, but using athomasson2/ebook2audiobookxtts:huggingface still does.

EDIT: Now it seems to work, maybe I tried just as it was being changed lol

[Kombustor]

I'm still having this issue in a duplicated Hugging face space. I tried rebuilding the docker image to use the newest commit but it still fails with Error: cannot unpack non-iterable async_generator object

[DrewThomasson]

I'm still having this issue in a duplicated Hugging face space. I tried rebuilding the docker image to use the newest commit but it still fails with Error: cannot unpack non-iterable async_generator object

The custom model issue should be fixed in the next PR for v2.0

If you want you can try using the legacy v1.0 (but its old were not fixing any bugs from it as were working into v2.0 now)

• But it should "kind of" work with the custom models for rn

https://github.com/DrewThomasson/ebook2audiobook/tree/v1.0

For the legacy v1.0 docker image it's

-Run with CPU only

docker run -it --rm -p 7860:7860 --platform=linux/amd64 athomasson2/ebook2audiobookxtts:huggingface_v1.0 python app.py

-Run with GPU Speedup (Nvida graphics cards only)

docker run -it --rm --gpus all -p 7860:7860 --platform=linux/amd64 athomasson2/ebook2audiobookxtts:huggingface_v1.0 python app.py

Also for the legacy v1.0 huggingface if you want to duplicate it

https://huggingface.co/spaces/drewThomasson/ebook2audiobook_v1.0

[dxcore35]

BryanCranston

Not sure if this is releveant, when using any fine-tuned voices the dowload is stuck on this:

[DrewThomasson]

BryanCranston

Not sure if this is releveant, when using any fine-tuned voices the dowload is stuck on this:

That seems to be just a huggingface issue tbh