Subjective Voice Quality of the IVAS Codec

Abstract

The 3GPP Immersive Voice and Audio Services (IVAS) codec enables spatial audio communications using either stereo, multi-channel, Scene-Based Audio (SBA), objects or Metadata-Assisted Spatial Audio (MASA) formats. Combinations such as objects together with SBA or MASA are also supported. This paper presents subjective listening test results conducted with extended Absolute Category Rating (ACR) Mean Opinion Score (MOS) test methodology. The main interest is the perceived subjective audio quality achieved with the introduction of the spatial audio to voice communications. Previous generation monaural 3GPP voice codecs are included in the same test, which allows audio quality improvements over the past decades to be studied. 3GPP codecs include AMR (narrowband), AMR-WB (wideband) and EVS. Spatial audio coding is conducted with the IVAS codec operating in stereo, ambisonic and MASA formats. All test signals originate from true spatial audio recordings, enabling both monaural and spatial representations to be generated from the same inputs. The full results are shown in both curves with increasing bitrate and bars with 95% confidence intervals.