Neural Imitation of Human Perceptual Response to High-Resolution Audio
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Computer audition systems based on neural processing provide deep perceptual understanding of digital audio signals but often rely on extensive manual annotation. In this paper, we propose an unsupervised framework to imitate human perceptual processing for high-resolution audio quality assessment. We employ a data format conversion method to quantize continuous audio waveforms into discrete token sequences. A Large Language Model is then utilized to model these sequences via next-token prediction, capturing complex temporal dependencies without labeled data. Higher prediction accuracy is observed on truly recorded high-resolution content, while artificially up-sampled content induces significant prediction drops. Experimental results demonstrate that this unsupervised approach effectively distinguishes between true and synthesized high-resolution audio. Manual verification result shows strong agreement with the results provided by the proposed neural model.
Fields of Study & Subjects
Music and Audio Processing
Speech and Audio Processing
Speech Recognition and Synthesis
Publication details
- Journal / source
- International Conference on Consumer Electronics
- Pages
- 1–4
- Publication date
- 3 February 2026
- Language
- EN