MASA 1Dir1TC at 48kbps systematically worse than 32kbps due to artifacts and bandwidth or noise-filling issues

Basic info

• Float reference:77d9fdad
• Fixed point:e55c584d

Bug description

There seems to be a systematic problem with MASA 1Dir 1TC at 48kbps which can cause it to sound worse than 32kbps and even 24.4kbps for the same input. Discovered for many items selected for BS1534-13. This is full chain fixed-point. The problem manifests mainly in what seems to be either a bandwidth/high-band scaling issue or a missing noisefilling (just from looking at the spectrograms, no debugging done yet), but there are also more "crispy" artifacts sometimes. Possibly those are only audible due to the missing noisefilling (if that is indeed the problem). Quality impact is very high, for me would be at least 10 Mushra points. Example below:

Fx at 32kbps:

Fx at 48kbps:

For comparison, here is float full chain at 48kbps which looks better:

Ways to reproduce

Get bitstream and input wav files from Box in issues/basop-issue-2014.

./IVAS_cod_fx -masa 1 lm13a1s06.wav.met 48000 48 lm13a1s06.wav bit
./IVAS_dec_fx BINAURAL 48 bit out.wav