Skip to content

Core synthesis issues in ISM 1

Basic info

Bug description

The fixed point encoder shows different artefact during active speech for ISM-1 at 16.4 kbps (can be observed at 24.4 and 32 kbps as well). This also means that all mode using the core at 16.4, 24.4 and 32 kbps ( or bitrates close to that) will suffer similar issues.

The first 2 images below show probably a transition artefact when switching from one core to another OR a scaling issue. In both cases the time domain signal exhibits a fast change of energy in the fixed point. The 3rd image shows that is way too noisy, even in the LF.

From top to bottom:

  • Floating point synthesis at 16.4 kbps
  • Fixed point synthesis at 16.4 kbps
  • Spectrums of both synthesis

image

image

image

Ways to reproduce

Input files:

G4S5.wav.0.csv

G4S5

IVAS_cod_flt -ISM 1 G4S5.wav.0.csv 16400 48 G4S5.wav bit.flt

IVAS_cod_fx -ISM 1 G4S5.wav.0.csv 16400 48 G4S5.wav bit.fx

IVAS_dec_flt EXT 48 bit.flt syn.flt.wav

IVAS_dec_flt EXT 48 bit.fx syn.fx.wav

(Clear steps or refer to a failing automated test, e.g. with a pipeline link)

Edited by vaillancour