FOfollower·May 29, 2024
The repo does claim that something is "open source" but the only included license text is "CC-BY-NC-ND" and the README seems to restrict permissions even further.
In addition, the Hugging Face repo[-1] states the license as "Creative Commons Attribution Non Commercial 4.0" (lacking the "ND").
Unfortunately, this combination of license imprecision and restrictiveness seems par for the course with a lot of academic TTS projects. (And, even for commercial "Open Source" TTS projects it's often the case that while code might be OSS licensed, none of the the voice data/models are.)
The current version[0] of the README repo states:
* "The open-source version on HuggingFace is a 40,000 hours pre trained model without SFT." (Presumably refers to model.)
* "At the same time, we have internally trained a detection model and plan to open-source it in
Read the rest
the future." (Not directly relevant.)
The included "Roadmap" indicates related completed & uncompleted tasks:
* "[x] Open-source the 40k hour base model and spk_stats file"
* "[ ] Open-source VQ encoder and Lora training code"
* "[ ] Open-source the 40k hour version with multi-emotion control"
However, as noted, the current LICENSE[1] file states:
* "Attribution-NonCommercial-NoDerivatives 4.0 International"
And the README also contradicts the license:
* "This repo is for academic purposes only. It is intended for educational and research use, and should not be used for any commercial or legal purposes."
* "The information and data used in this repo, are for academic and research purposes only."
And this part of the "disclaimer" would make me concerned about potential licensing issues in regard to code and or data from other sources:
* "The data obtained from publicly available sources, and the authors do not claim any ownership or copyright over the data."
The code in the repo itself appears to have no license information contained within it.
My go-to actually Open Source licensed Text-To-Speech project (with a range of voice[2] model licenses[3]--including Public Domain & CC-BY[4]) is Piper TTS: https://github.com/rhasspy/piper
---- footnotes ----
[-1] https://huggingface.co/2Noise/ChatTTS
[0] https://github.com/2noise/ChatTTS/blob/f4c8329f0d231b272b676...
[1] https://github.com/2noise/ChatTTS/blob/f4c8329f0d231b272b676...
[2] Voice samples: https://rhasspy.github.io/piper-samples/
[3] Though I would also caution that (at least by my interpretation) some of the voices listed as CC0/PD or CC-BY also note that they've been "fine-tuned" on models which have more restrictive licenses and thus probably can't inherit the voice data's more permissive license.
[4] However, these Public Domain and/or CC-BY voices do appear specifically created to be data license-compliant: https://brycebeattie.com/files/tts/