Source Separation for A Cappella Music

2025/09/30 by Lanzendörfer, Luca A., Pinkl, Constantin, Grötschla, Florian
#FOS: Computer and information sciences #Machine Learning (cs.LG) #Sound (cs.SD)

paper · doi:10.48550/arxiv.2509.26580

Abstract

In this work, we study the task of multi-singer separation in a cappella music, where the number of active singers varies across mixtures. To address this, we use a power set-based data augmentation strategy that expands limited multi-singer datasets into exponentially more training samples. To separate singers, we introduce SepACap, an adaptation of SepReformer, a state-of-the-art speaker separation model architecture. We adapt the model with periodic activations and a composite loss function that remains effective when stems are silent, enabling robust detection and separation. Experiments on the JaCappella dataset demonstrate that our approach achieves state-of-the-art performance in both full-ensemble and subset singer separation scenarios, outperforming spectrogram-based baselines while generalizing to realistic mixtures with varying numbers of singers.

Citations

• TF-Locoformer: Transformer with Local Modeling by Convolution for Speech Separation and Enhancement
• Separate and Reconstruct: Asymmetric Encoder-Decoder for Speech Separation
• Mel-Band RoFormer for Music Source Separation
• High-Fidelity Audio Compression with Improved RVQGAN
• A Phoneme-Informed Neural Network Model for Note-Level Singing Transcription
• MossFormer: Pushing the Performance Limit of Monaural Speech Separation using Gated Single-Head Transformer with Convolution-Augmented Joint Self-Attentions
• Hybrid Transformers for Music Source Separation
• MedleyVox: An Evaluation Dataset for Multiple Singing Voices Separation
• Music Source Separation with Band-split RNN
• RoFormer: Enhanced Transformer with Rotary Position Embedding
• Attention is All You Need in Speech Separation
• All for One and One for All: Improving Music Separation by Bridging Networks
• Deep Learning Based Source Separation Applied To Choir Ensembles
• Dual-Path Transformer Network: Direct Context-Aware Modeling for End-to-End Monaural Speech Separation
• Neural Networks Fail to Learn Periodic Functions and How to Fix It
• Demucs: Deep Extractor for Music Sources with extra unlabeled data remixed
• A Framework for Multi-f0 Modeling in SATB Choir Recordings
• SDR - half-baked or well done?

Related

• Adversarial Semi-Supervised Audio Source Separation applied to Singing\n Voice Extraction
• A cappella: Audio-visual Singing Voice Separation
• MedleyVox: An Evaluation Dataset for Multiple Singing Voices Separation
• Mamba2 Meets Silence: Robust Vocal Source Separation for Sparse Regions
• Deep Learning Based Source Separation Applied To Choir Ensembles