That is totally false. dada2 does not use any databases — it trains an error model based on the error rates observed in the run itself. I believe the original dada2 paper was benchmarked on bacterial community data, but there is no evidence to suggest that ITS data are going to exhibit different error rates on a sequencing run, requiring different methods.
dada2 is pretty widely used for ITS at this point. In addition to the tutorials that @Mehrbod_Estaki highlighted, you may also want to check out the q2-itsxpress tutorial, which uses q2-itsxpress to trim ITS reads prior to denoising with dada2: