Handle number on ASR

Hi there can anyone help me to find way to handle numbers on Speech recognition model, I’m working on some low resources language but sometime audio may contains number that
are spelling in french like 2000, 6000, ect
I’m trying to finetune MMS or Wav2vec2 on wolof where sometime audio may contains numbers
cc @patrickvonplaten

I used GitHub - savoirfairelinux/num2words: Modules to convert numbers to words. 42 --> forty-two for this. Supports many languages, as well as ordinal numbers and years