next steps after shutdown

[bachittle]

According to the main site https://coqui.ai/ , coqui is shutting down, which is unfortunate as these open source libraries are great and could still be maintained. I'm wondering if there are any next steps to proceed, like if the license should allow for commercial use and the open source community could fork this repository to keep it alive. Hoping we can still get use out of it because I'd say it is currently the best open-source voice synthesis and cloning toolkit out there at the moment.

For reference, I also have tried standalone bark, StyleTTS 2 and OpenVoice as alternatives, but I find the voice cloning was not as good as this library (general synthesis is pretty good, but cloning in particular is hard to get right).

[slavakurilyak]

+1 for clarity on commercial licensing terms

[NiluK]

Further info from tweet: https://twitter.com/_josh_meyer_/status/1742522906041635166

I can get why they're shutting down, Google, OpenAI, Elevenlabs, Amazon, etc TTS space is really crowded.

[SomeDevWeb]

O man, and I just restarted looking into coqui today and when I went online to refresh myself on some commands, I saw the website closed :( But just to say, I still have it installed on my Win laptop, but on its Linux (WSL) and most of the models still work. I even tested some new ones and they downloaded OK. Some ofcourse give errors, but that's expected. So ya, I surely hope the community will pick this up and reignite the fire! As @NiluK said, the author incourages people to continue the effort... :)

Hm... I was wondering if it would be a good idea to install a server somewhere and create a very simple web GUI so we can test it that way...

[alexiri]

The code is licensed MPL-2.0 license. Should be fine to fork and continue the project, no?

[jp-x-g]

Yeah, the thing I was confused about was whether there was any plan on the part of the repository's current owners to continue maintaining it. Per CODE_OWNERS.rst there's supposed to be someone to consult about stuff. @erogol @reuben @Edresson Do you all intend to continue as maintainers (accepting PRs etc) or should we make a new fork?

[erogol]

I don't plan to continue maintaining the code atm. things might change in the future.

[enzyme69]

Man that's unfortunate, so Rabbit AI is buying coqui XTTS? I love using it so simple and easy~

[metal3d]

You should take a look at the long discussion here https://github.com/coqui-ai/TTS/issues/3490

There's no aggression, just great disappointment and a desire on the part of the community to use XTTS without restriction. The aim is to promote the model and try to revive it, in the hope of relaunching it in a truly "free" mode.

Because, frankly, the situation is bitter. I was going to propose to my employers the use of this model for customers, they would have paid, and we on our side would have enormous satisfaction in offering a solution of this quality.

If the model becomes completely closed, resold to "who knows who", we won't use it... and we'll still be stuck with Google and Amazon APIs because they're cheaper (despite all the scandals they cause, our customers prefer these solutions because nobody else offers anything better).

I'm going to reiterate a comparison that I insist on heavily, but Blender suffered the same fate 20 years ago. Luckily, instead of selling the source code to a company that was going to close the software down, Ton Roosendaal had the genius to find a way of making Blender completely free to access and use. Today, it's one of the best 3D softwares on the market, and the Blender foundation earns enough money to pay salaries and recruit.

The "hybrid" mode of XTTS trained was not a wise choice. But please, let us use it commercially, and you can try to sell support and/or a studio or datasets, or API access. Blender does it, you can do it too.

[metal3d]

(this situation reminds me what Sonos made to the Snips project... we, users, have filled the dataset, and one day Snips has sold the entire project that became entirely closed... don't do the same, please)

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.

[eginhard]

FYI, we're continuing to maintain a fork of Coqui TTS: https://github.com/idiap/coqui-ai-TTS (PyPI package: coqui-tts)

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.

[Paillat-dev]

Nope

[metal3d]

FYI, we're continuing to maintain a fork of Coqui TTS: https://github.com/idiap/coqui-ai-TTS (PyPI package: coqui-tts)

That's very nice. But the trained model license is still a remaining question. CMPL models should not be used commercially, and we cannot buy license. If we “fine tune” the model, the license is unclear, because the model is not "in the form provided by the licensor" afterward.

I'm very happy to know that someone continues to maintain the library.

[eginhard]

That's very nice. But the trained model license is still a remaining question. CMPL models should not be used commercially, and we cannot buy license. If we “fine tune” the model, the license is unclear, because the model is not "in the form provided by the licensor" afterward.

IANAL, but the CPML is relatively clear and I don't see why derivative works would not have to adopt the same license, see also: "You must ensure that anyone who gets a copy of any part of the model, or any modification of the model, or their output, from you also gets a copy of these terms or the URL for them above."

The part you quoted only refers to patent claims, but in any case, to obtain a fine-tuned model you first have to use the original one "in the form provided".

[theo77186]

This now boils down to whoever gets the models' IP. If the IP got transferred to any entity, any relicensing is unlikely and CPML remains fully enforceable. However, if there is no successor entity that takes the IP, the legal status is becoming muddy. Copyright should still apply de jure in this case, but whether the copyright is enforceable at all isn't clear because of the lack of IP owner. This legal uncertainty is nonetheless enough to deter any commercial use of these models. But IANAL.

[metal3d]

Having no information from the authors in any case, despite all our injunctions, requests for information, requests for details, and given the CMPL's legal vacuum on the fact that a re-trained model (whether ve is fine-tune or not), I'm going on the assumption that there are only 3 possibilities:

• the author doesn't have the answer
• the author doesn't give a damn
• the author wants to sell the tool and thumb his nose at the community that provided support and/or paid for a license

Yes, I'm angry. Angry that the open-source philosophy is constantly insulted by this kind of method. For over 30 years, people have been trying to assert the right to share and freely use computer tools. For 30 years, people have been using this approach as a springboard for their own goals, namely: to stab you in the back at the last moment when you're pretty sure you can make money, even if it means betraying the people who support you.

I've always given a little money to foundations and royalty-free projects. I would have done it for CoquiTTS. But the CMPL worried me from the start. I was right not to give a cent!

Sorry for the rant.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.

[Paillat-dev]

Stale bots should be forbidden

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.

[Paillat-dev]

This is activity