Exploring Perceptual Audio Quality Measurement on Stereo Processing Using the Open Dataset of Audio Quality

TL;DR

This work analyzes the Open Dataset of Audio Quality (ODAQ) update focused on stereo processing (Mid/Side and Left/Right) to evaluate how existing objective metrics predict subjective quality across monaural and binaural distortions. It compares a range of timbre- and spatial-quality metrics, including MoBi-Q, eMoBi-Q, PEMO-Q, and binaural extensions of PEAQ, using MUSHRA tests and careful pre-processing to reveal strengths and limitations in predicting stereo degradation. The findings show timbre-based metrics often predict well in simple contexts, while binaural metrics offer limited gains and can be confounded by presentation context and hard-panned artifacts, highlighting the need for models that jointly capture bottom-up psychoacoustics and top-down listening factors. The results guide future metric development toward robust integration of timbral and spatial cues, with data-driven handling of context to improve perceptual quality prediction in realistic stereo scenarios.

Abstract

ODAQ (Open Dataset of Audio Quality) provides a comprehensive framework for exploring both monaural and binaural audio quality degradations across a range of distortion classes and signals, accompanied by subjective quality ratings. A recent update of ODAQ, focusing on the impact of stereo processing methods such as Mid/Side (MS) and Left/Right (LR), provides test signals and subjective ratings for the in-depth investigation of state-of-the-art objective audio quality metrics. Our evaluation results suggest that, while timbre-focused metrics often yield robust results under simpler conditions, their prediction performance tends to suffer under the conditions with a more complex presentation context. Our findings underscore the importance of modeling the interplay of bottom-up psychoacoustic processes and top-down contextual factors, guiding future research toward models that more effectively integrate both timbral and spatial dimensions of perceived audio quality.

Figures (7)

• Figure 1: Experimental setup for the generation and quality assessment of the degraded audio files.
• Figure 2: Experimental conditions for the degraded files. In the MUSHRA test, listeners were presented with LR-only degradations (applied independently to the left and right channels), MS-only degradations (applied to the mid and side channels), or a mixture of both.
• Figure 3: Correlation between objective metric predictions and subjective quality scores for each experiment, taking into account artifact type and presentation context. $\hbox{CI}_{95\%} \leq \pm 0.01$ for all estimates.
• Figure 4: Correlation between objective metric predictions and subjective quality scores for each audio excerpt and their respective treatments. Audio excerpts without hard-panned auditory objects. $\hbox{CI}_{95\%} \leq \pm 0.01$ for all estimates.
• Figure 5: Correlation between objective metric predictions and subjective quality scores for each audio excerpt and their respective treatments. Audio excerpts with hard-panned auditory objects. $\hbox{CI}_{95\%} \leq \pm 0.01$ for all estimates.