# Music Source Separation Based on a Lightweight Deep Learning Framework (DTTNET: DUAL-PATH TFC-TDF UNET)

arXiv: https://arxiv.org/abs/2309.08684 · HTML: https://mlsanity.com/paper/2309.08684/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 15 Sep 2023.

## Built on by (most cited first, 11 of 11)

- [A Two-Stage Band-Split Mamba-2 Network For Music Separation](https://mlsanity.com/paper/2409.06245/) (2024)
- [User-guided Generative Source Separation](https://mlsanity.com/paper/2507.01339/) (2025)
- [Remixing Music for Hearing Aids Using Ensemble of Fine-Tuned Source Separators](https://mlsanity.com/paper/2401.06203/) (2024)
- [The first Cadenza challenges: using machine learning competitions to improve music for listeners with a hearing loss](https://mlsanity.com/paper/2409.05095/) (2024)
- [An Ensemble Approach to Music Source Separation: A Comparative Analysis of Conventional and Hierarchical Stem Separation](https://mlsanity.com/paper/2410.20773/) (2024)
- [Training-Free Multi-Step Audio Source Separation](https://mlsanity.com/paper/2505.19534/) (2025)
- [The Spheres Dataset: Multitrack Orchestral Recordings for Music Source Separation and Information Retrieval](https://mlsanity.com/paper/2511.21247/) (2025)
- [Summary of The Inaugural Music Source Restoration Challenge](https://mlsanity.com/paper/2601.04343/) (2026)
- [A Conditioned UNet for Music Source Separation](https://mlsanity.com/paper/2512.15532/) (2025)
- [The Costs of Reproducibility in Music Separation Research: a Replication of Band-Split RNN](https://mlsanity.com/paper/2603.09187/) (2026)
- [DTT-BSR+: A Generative-Regression Cascade for Music Source Restoration](https://mlsanity.com/paper/2606.24127/) (2026)

## Abstract (arXiv)

Music source separation (MSS) aims to extract 'vocals', 'drums', 'bass' and 'other' tracks from a piece of mixed music. 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, 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.