A case for TWO!EARS in audio quality assessment

A Raake, H Wierstorf, J Blauert, "A case for TWO!EARS in audio quality assessment," in FORUM ACUSTICUM, (2014). [ paper ]

Bibtex

@inproceedings{Raake2014,
    title     = {{A case for TWO!EARS in audio quality assessment}},
    author    = {Raake, Alexander and Wierstorf, Hagen and Blauert, Jens},
    booktitle = {Forum Acusticum},
    address   = {Krakow, Poland},
    month     = {September},
    year      = {2014}
}

Abstract

The paper presents first results on audio-quality evaluation following the auditory-perception modeling paradigm of TWO!EARS, a FET-Open project funded under FP7 ICT by the European Commission (www.twoears.eu). The project targets an interactive system for binaural auditory perception, including input from the visual modality. In its ultimate form, it will be based on an interactive robot platform combining bottom-up cues from monaural and binaural signal processing with hypothesis-driven top-down cognition, organized around an expert-system with black-board architecture. One of the two proof-of-concept applications is that of audio-quality evaluation for loudspeaker-based reproduction. Here, targeted key innovations include a dedicated scene-based evaluation paradigm, active exploration of sound fields using head-movements and displacement, the combination of bottom-up information with top-down feedback and adaptation, and the dedicated use of internal rather than external references within the expert system. The paper presents research conducted in the course of model development. Two first findings along these lines will be presented: (1) A listening test was conducted to assess the localization performance of human listeners for several near-field compensated higher-order Ambisonics setups. A binaural model was able to predict the perceived direction with good accuracy. However, for some positions in the listening area, listeners reported to perceive more than one auditory event. This result was not predicted by the binaural model, which instead returned a single source direction. By endowing the binaural model with the ability of head rotations, it could be expanded so as to more accurately predict whether the listener perceived one or more sources, and from which direction. (2) A first pilot test has been conducted to address the scene-based test-paradigm targeted by TWO!EARS. Here, a scene with two guitars and one singer playing a musical piece was reproduced via different Wave Field Synthesis systems so as to selectively degrade one or multiple of the three sources. The different conditions were assessed in a paired comparison preference test, indicating, among other findings, that the reference scene was clearly not the preferred one. The paper summarizes the results and provides an outlook on future developments on sound quality modeling in the TWO!EARS project.