about dc_crn training

[Swimaster]

I trained dc_crn, and make the mode to "mapping". I remove the argument "--normalize_output_wav true" in "enh.sh", but in stage 7, the output wav is still scale-normalized. But this is not the case in other models whose mode is "masking" (conv-tasnet、dccrn). Why?

[sw005320]

Thanks for your question.
@Emrys365, can you answer this question?