Skip to content

The temporal excitation of the GSC could be wrong

Basic info

  • Float reference:
    • Encoder (float):e15ad0ad
    • Decoder (float):
  • Fixed point:
    • Encoder (fixed): ff4d7358
    • Decoder (fixed):

Bug description

When forcing IVAS to GSC and analysing the output of the GSC module, it was observed that the fixed point pitch was often too low and the energy of the time domain excitation was not consistent with the floating point The figure below compares the excitation after the module enc_pit_exc_ivas() when GSC_IVAS_mode is forced to 1. When looking at the spectrum (last plot) it is possible to observe the energy mismatch between float and fixed excitation.

image

That is not a major correction as these modes are rarely used, and the frequency domain is compensating the energy. but it forces the GSC frequency domain to use only a part of it gain quantizer.

Ways to reproduce

Box folder: ...\Box_EXTERNAL_IVAS_BASOP_VERIFICATION\issues\issue-xxxx

-stereo -force GSC 16400  48 C:\Temp\dp03a4s03.wav bit  (could be any file)