Multilingual multiple languages fine-tuning on facebook mms model

Models
1

Hi everyone,

I am trying to finetune facebooks mms models on some large internal data which i believe would lower the wer for these respective langauges even more.

The thing is I have over 15 diverse languages to fine-tune. but i am unsure on how to fine-tune them all on a single model.

The huggingface fine-tuning code says use an adapter for the given language and fine-tune it. should train the multiple languages sequential on per language basis ? like iterating over every lang_dataset and lang_adapter to fine-tune them all sequential ?

If someone knows any possible solution on how to go about with this please let me know :pray:

2

gently pinging @sanchit-gandhi

3

hey , you cant just fine tune all of 15 languages on a single model , i believe you should use differennt models that come under facebook-mms-tts for every different language .Whichever model is suitable for a particular language .there are almost 1000 + languages supported by facebook-mms-tts and for various languages there are sub models of facebook-mms-tts that you can use .