Hi @MiriamGorostidi,
if the sequences are from IonTorrent, you are right in say that the reads are single reads but the error profile is different from the Illumina reads so you should use: the dada2 denoise-pyro denoise-pyro: Denoise and dereplicate single-end pyrosequences — QIIME 2 2020.8.0 documentation
For training your database I suggest to use the brand new Rescript plugin: Processing, filtering, and evaluating the SILVA database (and other reference sequence data) with RESCRIPt
As alternative, regarding trying the clustering approach, please look at:
https://docs.qiime2.org/2020.8/tutorials/otu-clustering/
In the 'Closed-reference clustering' part. For this you don't need to train the database.
Hope it helps