SBA + Binaural Headrotation: Trajectories with unnormalized quaternions have impact on loudness
Basic info
- Commit SHA: b4edaaa8
- Platform: float; BASOP not tested
Bug description
When decoding Ambisonics for the 3 and 4 transport channel configurations, feeding an hard rotation trajectory which is not normalized to 1 into the decoder seems to have a massive impact on the loudness of the binaural signal.
This is the trajectory in question:
0.00000,-3.000000,0.000000,0.000000| Trajectory | Loudness (bs1770) |
|---|---|
| no trajectory | -18.418470 LKFS |
| Unnormalized trajectory (see above) | -12.138492 LKFS |
| full-circle-15s (for comparison) | -18.862067 LKFS |
This doesn't happen at the the 1 and 2 TC configurations (e.g. 32 and 48 kbps). Also, there is no significant amplification/attenuation using e.g. the full-circle-15s trajectory.
I know this not how the trajectory actually should look like, but this behavior is still not good. If this can't be fixed algorithically, the API should make sure that only a valid trajectory is handed down to the rotation routine.
Ways to reproduce
Bitstream: stvFOA48c_128kbps.192
Trajectory: trajectory_test.csv
IVAS_dec -T ./trajectory_test.csv binaural 48 stvFOA48c_128kbps.192 out.wav