Request for Clarification and Update Regarding OmniVoice Model License for Commercial Use
Dear Team,
I hope you are doing well.
We are currently evaluating and integrating OmniVoice into our speech technology platform, and we are interested in understanding the licensing terms for commercial use of the pre-trained model.
While reviewing the licensing information, we noticed a potential discrepancy between the Hugging Face repository and the official GitHub repository.
1. Hugging Face repository
In the OmniVoice Hugging Face repository, the license section states:
Our code is released under the Apache 2.0 License. The pre-trained model is licensed under the CC-BY-NC due to constraints from its training data (e.g., Emilia).
Based on this statement, it appears that the source code is available under Apache-2.0, while the pre-trained model weights may be subject to the CC-BY-NC license. Since CC-BY-NC generally restricts commercial use, this raises concerns about whether the pre-trained model can legally be used in a commercial product or service.
2. GitHub repository
However, in the official OmniVoice GitHub repository, the project is presented as commercially usable under the Apache-2.0 License.
Additionally, https://huggingface.co/datasets/amphion/Emilia-Dataset As per this link Emilia data set is now cc-by-4.0 not cc-by-nc.
This creates uncertainty about the actual licensing terms applicable to the pre-trained model weights, as opposed to the source code.
Could you please help us clarify the following points?
Is the OmniVoice pre-trained model available for commercial use under the Apache-2.0 License, or is it still subject to the CC-BY-NC license mentioned on Hugging Face?
If the pre-trained model weights are indeed commercially usable under Apache-2.0, could you please confirm this explicitly and update the license information in the Hugging Face repository accordingly? An official clarification in the model card or a separate license file would help users and organizations make informed decisions.
Thank you for your time and support. We look forward to your clarification.
Hi, the model weights are non-commercial and can’t be used commercially.
Thank you for the clarification.
Just wanted to understand a little more about why the model is non-commercial .
We noticed that the Emilia dataset is now listed under the CC-BY-4.0 license, which permits commercial use:
Emilia Dataset on Hugging Face
We have also checked the other primary datasets mentioned in your white paper, and our understanding is that they are available under licenses such as Apache-2.0, which also permit commercial use.
Could you please clarify which specific dataset, data source, or licensing restriction requires the OmniVoice pre-trained model weights to remain non-commercial?
This would help us better understand the licensing basis for the model weights and determine whether there are any specific restrictions we need to consider for commercial deployment.
Thank you!
Emilia has two parts: one under CC-BY-4.0, the other under CC BY-NC 4.0. Many other datasets in our training set, such as WenetSpeech-Yue, also use non-commercial licenses.
Another piece to note about commercial licensing is that there is another transitive dependency on higgs audio v2 tokenizer which has its own license that requires additional permissions for any product with more than 100k annual users.
https://huggingface.co/bosonai/higgs-audio-v2-tokenizer/blob/main/LICENSE