Predicting Subjective Ratings of Guitar Amplifier Emulations

Wright A, Dal Santo G, Schlecht SJ. Predicting Subjective Ratings of Guitar Amplifier Emulations. J Audio Eng Soc. 2026;74(10):762-772. doi:10.17743/jaes.2026.0304

@article{Wright2026_23442,
  author = {Wright, Alec and Dal Santo, Gloria and Schlecht, Sebastian J.},
  title = {{Predicting Subjective Ratings of Guitar Amplifier Emulations}},
  journal = {Journal of the Audio Engineering Society},
  volume = {74},
  number = {10},
  pages = {762--772},
  year = {2026},
  month = oct,
  publisher = {Audio Engineering Society},
  doi = {10.17743/jaes.2026.0304},
  url = {https://doi.org/10.17743/jaes.2026.0304}
}

TY  - JOUR
TI  - Predicting Subjective Ratings of Guitar Amplifier Emulations
AU  - Wright, Alec
AU  - Dal Santo, Gloria
AU  - Schlecht, Sebastian J.
T2  - Journal of the Audio Engineering Society
J2  - J. Audio Eng. Soc.
VL  - 74
IS  - 10
SP  - 762
EP  - 772
PY  - 2026
DA  - 2026/10/09
DO  - 10.17743/jaes.2026.0304
UR  - https://doi.org/10.17743/jaes.2026.0304
PB  - Audio Engineering Society
LA  - en
AB  - Evaluating digital audio effects models of guitar amplifiers and distortion effects typically involves comparing emulations with reference recordings using objective metrics. Many widely used metrics, however, are poorly aligned with human perception. The authors construct a curated dataset of 590 emulation/reference audio pairs from previously published research on virtual analog modeling, with modeling targets including guitar amplifiers, distortion effects, compressors, and nonlinear filters. Using a subset of this dataset, a listening test is conducted to evaluate which reference-based audio metrics best predict perceptual similarity. It is found that time-frequency-based metrics, such as mel-frequency cepstral coefficients and deep feature-embedding distances exhibit a strong correlation with human perception. Conversely, it is found that commonly used time domain metrics, such as mean absolute error and error-to-signal ratio, perform poorly. With this, a clearer understanding of which metrics to use when reporting emulation results is provided and a simple model for predicting subjective similarity ratings from objective metrics is proposed.
ER  -