Results obtained for the test sets of the various experimental conditions (Music on the left side, and Speech on the right).

The performance of the various approaches is quantified using the root-mean-squared-error (rmse), Pearson’s linear correlation coefficient r, and the concordance correlation coefficient ccc. For each experiment, the performance measures were averaged across the five best trials. ID: intra-domain (M2M and S2S, baseline models trained and tested on the same database); CD: cross-domain (S2M and M2S); CDTL, cross-domain transfer learning (S2M and M2S); DS ⇒ DT (where DS is the source domain, and DT the target domain).