Is the model multilingual?
Is the model multilingual and how well does it perform on different languages?
best! wonderful!! For chinese an english
what about other languages? can you add benchmark results to model card?
Our model was trained on over 50 languages and delivers strong multilingual performance.
thank you for answering, i will try it when i have time, currently I am very busy with my own work. You could create space on HF for fast testing.
This deployment is based on our previous version, so the performance may be slightly worse than running inference directly with Hugging Face.
Wow you're so fast. I tried it on some Serbian podcast and in 60sec video it made 2 wrong words. One word was added that was not in audio and one word was guessed right but it was half in latin half in cyrillic script.
When training on Serbian datasets you should normalize scripts before training (choose one script, i prefer cyrillic but you can use either script). Also Croatian, Bosnian and Montenegrian are the same language except for a small set of words and different accent, so you probably could pretrain model on them, just be aware to normalize output to one script.
Thank you for your feedback. We’ll look into this issue carefully.