[Regression] Decoder Higher outlier for MASA 2TC 2Dir, Binaural Rendering

Basic info

• Float reference:
  • Encoder (float): deee9b97
  • Decoder (float): deee9b97
• Fixed point:
  • Encoder (fixed): n/a
  • Decoder (fixed): e2add3f0

Bug description

For an LTV-derived Testvector I'm seeing currently a higher max-diff difference between BASOP and Float than I've seen in April (April 23rd, ID 8ee4fc15)

• Max. abs. diff April 23rd: 646
• Max. abs. diff July 11th: 1531

I'm seeing this also for some neighbored bitrates, and it's not the case for EXT output:

	July 11th, 2025	July 11th, 2025	April 23rd, 2025	April 23rd, 2025
	Max. Abs. Diff EXT	Max. Abs. Diff Binaural	Max. Abs. Diff EXT	Max. Abs. Diff Binaural
ltv48_HOA2_MASA2tc2dir_64kbps	797	801	785	671
ltv48_HOA2_MASA2tc2dir_96kbps	697	1393	730	893
ltv48_HOA2_MASA2tc2dir_160kbps	427	1531	429	646
ltv48_HOA2_MASA2tc2dir_256kbps	161	1315	162	666
ltv48_HOA2_MASA2tc2dir_512kbps	24	544	27	863

Diff, FX - Float, April 23rd:

Diff, FX - Float, July 11th:

We can see that there is now significantly less noise, but there are these few "peaks" in the diff signal, which now stand out.

Ways to reproduce

Bitstream (zipped): ltv48_HOA2_MASA2tc2dir_160kbps.zip

IVAS_dec binaural 48 ltv48_HOA2_MASA2tc2dir_160kbps.192 out.wav