Text this: Nested U-Net-Based Speech Enhancement with Multi-Scale Feature Extraction and Dual-Path Time-Frequency Feature Modeling