Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces DTT-BSR, a GAN-based architecture for music source restoration that addresses both source separation and signal reconstruction from mixed and mastered recordings. It combines a RoPE transformer for capturing long-term dependencies with a dual-path band-split RNN for multi-resolution spectral processing. The model achieved competitive results in the ICASSP 2024 MSR Challenge, balancing performance with a relatively small parameter size (7.1M).
A hybrid GAN architecture with RoPE transformers and dual-path RNNs achieves state-of-the-art music source restoration with only 7.1M parameters.
Music source restoration (MSR) aims to recover unprocessed stems from mixed and mastered recordings. The challenge lies in both separating overlapping sources and reconstructing signals degraded by production effects such as compression and reverberation. We therefore propose DTT-BSR, a hybrid generative adversarial network (GAN) combining rotary positional embeddings (RoPE) transformer for long-term temporal modeling with dual-path band-split recurrent neural network (RNN) for multi-resolution spectral processing. Our model achieved 3rd place on the objective leaderboard and 4th place on the subjective leaderboard on the ICASSP 2026 MSR Challenge, demonstrating exceptional generation fidelity and semantic alignment with a compact size of 7.1M parameters.