k2-fsa

/

OmniVoice

Like 1.42k

Follow

k2-fsa 750

Thanks for the information. We have updated the model’s license to CC-BY-NC.

@zhu-han
I wonder if there will be a version that excludes Emilia? (Emilia-Yodas is CC-BY and can be used commercially)

@mrfakename

Hi, thanks for the reminder! I have just updated the gated repo prompt; you are free to use the Yodas subset under cc-by.

For the second inquiry, since the Emilia-Yodas and the original Emilia datasets are similar in nature, we have decided to maintain a single repository. If higher-quality or more finely annotated data becomes available in the future, we will consider creating a new repo.

Please note that Emilia-Yodas can also be loaded separately within this repo. For example:

from datasets import load_dataset

path = "Emilia-YODAS/DE/*.tar"
dataset = load_dataset("amphion/Emilia-Dataset", data_files={"de": path}, split="de", streaming=True)
print(dataset)  # This should only show 90 n_shards
print(next(iter(dataset['train'])))

Hello k2-fsa team,

I tested OmniVoice for Central Kurdish (Sorani), and the quality is excellent.

I am building a commercial education application and I would like to use OmniVoice for text-to-speech for paying users.

I understand that:

• the OmniVoice code is Apache 2.0
• the current pretrained model weights are CC-BY-NC

I would like to ask:

1. Do you offer a commercial license for the current OmniVoice pretrained weights?
2. If yes, how can I obtain or purchase that license?
3. Are you planning to release a commercial-safe checkpoint trained only on commercially usable datasets?
4. If a commercial version is planned, is there an estimated release timeline?

I am specifically interested in using OmniVoice for Kurdish Sorani TTS in a production application.

Thank you.