How to improve noises performance in ASR?

I had use AIshell-1 train an ASR model
But I find out that it had bad performance when i give some audio noise as input
(I except to get blank as output, but it’s not)
I try to add some noises as training data, but i don’t know how to label it
it will get error when i just give blank space as labeled

My question is

  1. Can i labeled blank ?
  2. Any other solution can improve background noises as input?

Hey, I don’t understand your request with 1. For 2. there is a whole literature on this subject, ranging from enhancing the signal with speech enhancement to robust ASR.

OK!! thank you for the replying