BASOP encoder has a large difference signal in noise signal using MASA selection items with DTX

Basic info

• Float reference:
  • Encoder (float): b63c033f
  • Decoder (float): b63c033f
• Fixed point:
  • Encoder (fixed): a6ac7971
  • Decoder (fixed): a6ac7971

Bug description

There is a bit peculiar large diff part when observing a selection test MASA item with purely noise content. See the figure below showing from left to right: full float, BASOP enc to float dec, and diff signal. Notably, if listening separately, the float and BASOP version are ok but the diff signal itself sounds almost the same as the versions in this 8.5 to 10 second period. This is also probably a source of pretty high MLD of 13.6 when the float enc to BASOP dec for this specific operation is almost bit exact. I am not assigning that high priority as this is kind of not audible but still this could point to some critical source of difference.

Ways to reproduce

Box folder: ...\Box_EXTERNAL_IVAS_BASOP_VERIFICATION\issues\issue-1498

Use the input file from Box with following command line

IVAS_cod -masa 2 BASOP_Selection_MUSHRA-7ad_s15.met -max_band fb -dtx 80000 48 BASOP_Selection_MUSHRA-7ad_s15.wav test_1571.bs
IVAS_dec EXT 48 test_1571.bs 1571_BASOP_Selection_MUSHRA-7ad_s15.wav_80k_48-48___EXT.wav