Speech Recognition with Large Pretrained Speech Models for Under-represented Languages