Music source separation (MSS) aims to extract a variety of sources from a piece of mixed music. Typically, in the context of MUSDB-18 demixing challenge, the target sources are ’vocals’, ’drums’, ’bass’ and ’other’ tracks. While deep learning methods have shown impressive results, there is a trend toward larger models. In our paper, we introduce a novel and lightweight architecture called DTTNet 1 , which is based on Dual-Path Module and Time-Frequency Convolutions Time-Distributed Fully-connected UNet (TFC-TDF UNet). DTTNet achieves 10.12 dB cSDR on ’vocals’ compared to 10.01 dB reported for Bandsplit RNN (BSRNN) but with 86.7% fewer parameters. We also assess pattern-specific performance and model generalization for intricate audio patterns.