Nick Zacharov

Andros Tjandra, Yi-Chiao Wu, Baishan Guo et al.

The quantification of audio aesthetics remains a complex challenge in audio processing, primarily due to its subjective nature, which is influenced by human perception and cultural context. Traditional methods often depend on human listeners for evaluation, leading to inconsistencies and high resource demands. This paper addresses the growing need for automated systems capable of predicting audio aesthetics without human intervention. Such systems are crucial for applications like data filtering, pseudo-labeling large datasets, and evaluating generative audio models, especially as these models become more sophisticated. In this work, we introduce a novel approach to audio aesthetic evaluation by proposing new annotation guidelines that decompose human listening perspectives into four distinct axes. We develop and train no-reference, per-item prediction models that offer a more nuanced assessment of audio quality. Our models are evaluated against human mean opinion scores (MOS) and existing methods, demonstrating comparable or superior performance. This research not only advances the field of audio aesthetics but also provides open-source models and datasets to facilitate future work and benchmarking. We release our code and pre-trained model at: https://github.com/facebookresearch/audiobox-aesthetics

Randy Frans Fela, Nick Zacharov, Søren Forchhammer

In an earlier study, we gathered perceptual evaluations of the audio, video, and audiovisual quality for 360 audiovisual content. This paper investigates perceived audiovisual quality prediction based on objective quality metrics and subjective scores of 360 video and spatial audio content. Thirteen objective video quality metrics and three objective audio quality metrics were evaluated for five stimuli for each coding parameter. Four regression-based machine learning models were trained and tested here, i.e., multiple linear regression, decision tree, random forest, and support vector machine. Each model was constructed using a combination of audio and video quality metrics and two cross-validation methods (k-Fold and Leave-One-Out) were investigated and produced 312 predictive models. The results indicate that the model based on the evaluation of VMAF and AMBIQUAL is better than other combinations of audio-video quality metric. In this study, support vector machine provides higher performance using k-Fold (PCC = 0.909, SROCC = 0.914, and RMSE = 0.416). These results can provide insights for the design of multimedia quality metrics and the development of predictive models for audiovisual omnidirectional media.

Randy Frans Fela, Nick Zacharov, Søren Forchhammer

This paper studies the quality of multimedia content focusing on 360 video and ambisonic spatial audio reproduced using a head-mounted display and a multichannel loudspeaker setup. Encoding parameters following basic video quality test conditions for 360 videos were selected and a low-bitrate codec was used for the audio encoder. Three subjective experiments were performed for the audio, video, and audiovisual respectively. Peak signal-to-noise ratio (PSNR) and its variants for 360 videos were computed to obtain objective quality metrics and subsequently correlated with the subjective video scores. This study shows that a Cross-Format SPSNR-NN has a slightly higher linear and monotonic correlation over all video sequences. Based on the audiovisual model, a power model shows a highest correlation between test data and predicted scores. We concluded that to enable the development of superior predictive model, a high quality, critical, synchronized audiovisual database is required. Furthermore, comprehensive assessor training may be beneficial prior to the testing to improve the assessors' discrimination ability particularly with respect to multichannel audio reproduction. In order to further improve the performance of audiovisual quality models for immersive content, in addition to developing broader and critical audiovisual databases, the subjective testing methodology needs to be evolved to provide greater resolution and robustness.