Table 8 \(95\,\%\) confidence intervals of the mean absolute error of the CNN across the fivefold cross-validation of the parameter extraction and all volumes. Results are reported for all time-frequency representations and the combination of distortion, tremolo and slapback delay effect
(a) Distortion + tremolo + slapback delay | ||||||
|---|---|---|---|---|---|---|
Gain [\(10^{-2}\)] | Tone [\(10^{-2}\)] | Depth [\(10^{-2}\)] | Frequency [\(10^{-2}\)] | Time [\(10^{-2}\)] | Mix [\(10^{-2}\)] | |
CNN + chromagram | 7.8 ± 0.12 | 13.4 ± 0. 16 | 11.8 ± 0.27 | 10.4 ± 0.46 | 19.4 ± 0.22 | 13.5 ± 0.22 |
CNN + GFCCs | 3.6 ± 0.06 | 4.9 ± 0.17 | 7.3 ± 0.14 | 9.1 ± 0.17 | 13.6 ± 0.18 | 8.0 ± 0.33 |
CNN + MFCCs | 3.5 ± 0.12 | 4.5 ± 0.13 | 9.0 ± 0.3 | 10.2 ± 0.44 | 13.3 ± 0.13 | 8.6 ± 0.1 |
CNN + spectrogram | 3.3 ± 0.014 | 4.1 ± 0.45 | 8.5 ± 0.3 | 8.6 ± 0.28 | 14.4 ± 0.12 | 10.3 ± 0.29 |