search

sh-lee-prml / HierSpeechpp

The official implementation of HierSpeech++
MIT License
1.18k stars 134 forks source link

Are you planning to add support for the Ukrainian language? #39

Open Danyilium opened 8 months ago

Danyilium commented 8 months ago

Dear developers,

I am asking you to add support for the Ukrainian language to your neural network HierSpeech++, which clones voices

Ukrainian is the native language of more than 40 million people worldwide, and adding it to your neural network will have a significant impact on the Ukrainian community.

Here are some of the reasons why I believe that the addition of the Ukrainian language will be useful:

 This will empower people with visual impairments and disabilities.
 This will make your neural network more accessible to people from all over the world who have Ukrainian roots.

I understand that this can be a difficult task, but I believe that it is worth it.

I would also like to mention that there is a large amount of Ukrainian language data that you can use to train your neural network. This data is publicly available from various sources

I believe that adding the Ukrainian language to your HierSpeech++ neural network is the same for the Ukrainian community. Thank you for your time and attention.

patriotyk commented 8 months ago

Unfortunately, there is no good Ukrainian tts datasets publicly available. I am working on one, but again I am doing it from books that might be copyrighted. So, first we need to have a good dataset. You can join our community on Discord : https://discord.gg/yVAjkBgmt4 or in the telegram https://t.me/speech_synthesis_uk and suggest your ideas or help here.

sh-lee-prml commented 8 months ago

Thanks for your interests

We found that it is difficult to evaluate the model with other language because we do not understand that language.

In addition, it is very hard to select the text-normalization method and phonemizer.

So, we have released the TTV trainer for people who want to train the model with their own dataset.

For our next plan, we will have a plan to release the GPT-based TTV model which can generate the speech with more realistic prosody in May.

Thanks!