ISM2 at 64kbps performs worse than at 48 kbps

The Listening test in Stage 2 (https://forge.3gpp.org/rep/ivas-codec-pc/ivas-codec/-/wikis/uploads/2f7d30dec8a610756901bf0a09ba3cd9/stage2_P800_7_ISM_2.pdf) revealed an issue in ISM2 at 64 kbps performing worse than at 48 kbps in clean speech.

A closer look points out to the segments where one of the two objects is an active speech while the other one is close to silence and thus coded by the low-rate core-coder mode. Consequently, the majority of the ivas_total_brate is assigned to code the active object and coded by the TCX core. The problem is that the total_brate in these segments diverges from nominal_brate used to configure the TCX core too much making a discrepancy between total_brate and nominal_brate which sounds very bad in the HF part of the spectrum.