This Corrigendum addresses P.862.2 systematic under-prediction of subjective scores. The under-prediction, 0.8 MOS on average, is due to the audio signals being exposed at an incorrect level to the loudness model. The issue leads to degradations being exaggerated and producing lower scores than expected.
The P.862.2 reference implementation provided in Annex A should be corrected to adjust the signals to a target level of 76 dB(SPL) for wideband assessment:
pesqmain.c, line 421:
float WB_InIIR_Hsos_8k[LINIIR] = { 0.251188 * 2.6657628f, 0.251188 * -5.3315255f, 0.251188 * 2.6657628f, -1.8890331f, 0.89487434f };
pesqmain.c, line 424:
float WB_InIIR_Hsos_16k[LINIIR] = { 0.251188 * 2.740826f, 0.251188 * -5.4816519f, 0.251188 * 2.740826f, -1.9444777f, 0.94597794f };
The updated conformance test file for P.862.2 is included in this Corrigendum (suppl23_wb.txt).
|
|
|
|
|