Assessing speech quality in noisy environments remains a significant challenge for research and development. This study evaluates the robustness of four objective models3QUEST, VISQOL, NISQA, and DNSMOS P.835by comparing their predictions with controlled listening test data from 32 participants. Three perceptual dimensions were analyzed: overall quality (G-MOS), speech quality (S-MOS), and noise intrusiveness (N-MOS). Results show that objective metrics reliably capture overall quality trends across diverse background noise scenarios, with correlations often exceeding 0.9. However, ranking system performance proved less consistent, particularly in complex conditions. Intrusive metrics demonstrated more substantial alignment with speech quality judgments. Higher-order ambisonics noise scenes did not weaken metric reliability. These findings highlight the usefulness of objective metrics for broad evaluation while emphasizing the need for hybrid approaches that incorporate targeted perceptual validation.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=23031
(532KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
