BASOP encoder: audible differences W channel in high-BR OSBA LTV at 256 kbps

Basic info

• Float reference:
  • Encoder (float):f7a0e207ee3622eb02c14e84586485b5795d912f
  • Decoder (float):f7a0e207ee3622eb02c14e84586485b5795d912f
• Fixed point:
  • Encoder (fixed):d74de338

Bug description

At some locations in the LTV for 2 ISM 3OA OSBA, there are large differences in the W signal. These are audible for instance at the following times in the signal.

• 42 s: the chirp frequency and the noise timbre sound clearly different between the float and BASOP encoders
• 1:29 s: the shakers miss HF energy and are less crisp in BASOP
• 1:41 s: differences in the HF energy lead to a slightly different timbre of the noise
• 2:29 s missing energy in the higher harmonics make the BASOP signal sound duller
• 2:31 s: HF energy missing in the noise in BASOP

Ways to reproduce

Run these commands with the BASOP and float encoders and the float decoder and compare the results between the two encoders.

 ./IVAS_cod -ism_sba 2 3 ../ivas-pc-testfiles/ltv48_OSBA_2ISM_HOA3_ISM1.csv ../ivas-pc-testfiles/ltv48_OSBA_2ISM_HOA3_ISM2.csv  256000 48 ../ivas-pc-testfiles/ltv48_OSBA_2ISM_HOA3.wav bit
../ivas-codec/ivas-float-update/IVAS_dec EXT 48 bit dec.wav

To find the differences, it is helpful to look at the differences of the W channels from both encoders. In the spectrograms of the difference signal, the strongest differences occur at high frequencies.

There may be a connection to the pure-SBA low-BR version of this (#1412 (closed)). The artifacts sound similar.