Background: We previously demonstrated that the high heterogeneity of response to computerized Auditory Training (AT) in psychosis can be ascribed to individual differences in sensory processing efficiency and neural plasticity. In particular, we showed that Auditory Processing Speed (APS) serves as a behavioral measure of target engagement, with faster speed predicting greater transfer effects to untrained cognitive domains. Here, we investigate whether the ability of APS to function as a proxy for target engagement is unique to AT, or if it applies to other training interventions, such as Executive Functioning Training (EFT). Additionally, we examine whether changes in APS are durable after these two forms of training. Methods: One hundred and twenty-five participants with Recent Onset Psychosis (ROP) were randomized to AT (n = 66) and EFT (n = 59), respectively. APS was captured at baseline, after treatment, and at 6-month follow-up. Mixed models repeated measures analysis with restricted maximum likelihood was used to examine whether training condition differentiated APS trajectories. Within-group correlational analyses were used to study the relationship between APS and performance improvements in each of the training exercises. Results: The two groups were matched for age, gender, education, and baseline APS. Participants showed high inter-individual variability in APS at each time point. The mixed model showed a significant effect of time (F = 5.99, p = .003) but not a significant groupby-time effect (F = .73, p = .48). This was driven by significant APS improvements AT patients after treatment (d = .75) that were maintained after 6 months (d = .63). Conversely, in EFT patients, APS improvements did not reach statistical significance after treatment (p = .33) or after 6 months (p = .24). In AT patients, baseline APS (but not APS change) highly predicted peak performance for each training exercise (all r's >.42).