Twenty-two listeners evaluated the perceptual similarity of three commercially available emulations of a Vox AC30 amplifier to the original hardware reference under controlled conditions: (1) a machine-learning plugin trained on the exact test amplifier, (2-3) two non-ML plugins modeling unspecified AC30 units. Using MUSHRA methodology, results showed the unit-specific ML plugin achieved perceptual indistinguishability from the hardware reference in 1/3 cases and outperformed both non-ML alternatives in 5/6 pairwise comparisons. This demonstrates that while non-ML plugins necessarily generalize across hardware units, ML techniques can achieve high perceptual fidelity when trained on target amplifiers. This represents a previously unattainable capability with significant implications for audio preservation and production.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=23019
(865KB)
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.
