People will presumably quite quickly start discussing events. Score data can be inferred from that.
But don’t know how hard it would be to process the data.
Put multiple different sources through multiple different LLMs and take a consensus of them. If there isn't a clear winner, declare it uncertain.
It'd be a bit of a bodge, but it'd work.
If you can get a live feed of the broadcast, you could also try feeding visual information in to be processed by a multimodal LLM.
Have the cricket score site open in a VM or something. Then a webjob can take screenshots and process the screenshot to extract the score.