@mukherjeesougata2399 - For svara v1, We trained it first on all the compiled data we had - including SYSPIN, RASA and others -- and then filtered it by UTMOS quality in subsequent rounds of training. Beyond that, no other data preparation steps were taken.
Lots of learning here, we're developing a much more thorough data control pipeline for svara v2, will keep update you when it's out :)
