Text this: Effect of Noises and GSM Codings on Pre-Trained Speaker Embedding Models in Forensic Voice Comparison