Text this: A Unified Deep Learning Framework for Short-Duration Speaker Verification in Adverse Environments