Spaces:
Restarting
Tannemirt
Thank you for creating this! I just played around a bit with the Tachelhit model with some sample words and phrases. It seems a bit inconsistent one the t, e (schwa), r, and g sounds. Are you planning to train new models? Would love to use this for my Tachelhit word of the day tweets.
You're most welcome. Just a clarification: I created the interface on Hugging Face but the models were created by Meta. While I would like to train new models, I currently don't have access to enough data. We've just started collecting some on Common Voice.
The models were trained on very little data (except Kabyle). And so they're not of very high quality. I recommend using the Kabyle model if you can. Also, note that the models were trained on text written in a particular (latin) script and orthography. Try to follow it to get the best results. Meta didn't release the training data but this is the closest thing I found so far.
If I ever publish new models, you'll find them here.
Cheers.
Azul, first of all thank you for all your hard work to make Tamazight has bigger community and easier to work with.
I'm just asking why I can't find Tamazight there is just dialect such Tarifit and tchlehit...
You're welcome! I do what I can.
I'm not sure what you mean here by "Tamazight". Any language has dialects and variants. If I type a word in English in Google Search to find out how it's pronounced I get a British pronunciation. If I wanted an American one I would have to find a model/service trained on American English.
So I don't know what you mean by "just dialects". In speech there are only dialects.