Online Independent Low-Rank Matrix Analysis as a Lightweight and Trainable Model for Real-Time Multichannel Music Source Separation

Taishi Nakashima, Nobutaka Ono
Proceedings of the AAAI 2026 Workshop on Audio-Centric AI: Towards Real-World Multimodal Reasoning and Application Use Cases (Audio-AAAI), PMLR 312:48-60, 2026.

Abstract

In this paper, we propose an online extension of independent low-rank matrix analysis (ILRMA) for blind music source separation under real-time constraints. Because multitrack stems are rarely released, we target lightweight processing that operates directly on in-the-wild mixtures. The method combines an online Itakura–Saito nonnegative matrix factorization (NMF) update with an online auxiliary-function independent vector analysis (IVA) framework, preserving the low-rank spectral model employed in ILRMA while updating the demixing matrix frame by frame with bounded latency and memory. Simulations on multitrack music mixtures show improved separation accuracy and a real-time factor below one, indicating feasibility for live and interactive scenarios. These results suggest blind separation suitable for low-latency music applications.

Cite this Paper

BibTeX
@InProceedings{pmlr-v312-nakashima26a, title = {Online Independent Low-Rank Matrix Analysis as a Lightweight and Trainable Model for Real-Time Multichannel Music Source Separation}, author = {Nakashima, Taishi and Ono, Nobutaka}, booktitle = {Proceedings of the AAAI 2026 Workshop on Audio-Centric AI: Towards Real-World Multimodal Reasoning and Application Use Cases (Audio-AAAI)}, pages = {48--60}, year = {2026}, editor = {Komatsu, Tatsuya and Imoto, Keisuke and Gao, Xiaoxue and Ono, Nobutaka and Chen, Nancy F.}, volume = {312}, series = {Proceedings of Machine Learning Research}, month = {26 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v312/main/assets/nakashima26a/nakashima26a.pdf}, url = {https://proceedings.mlr.press/v312/nakashima26a.html}, abstract = {In this paper, we propose an online extension of independent low-rank matrix analysis (ILRMA) for blind music source separation under real-time constraints. Because multitrack stems are rarely released, we target lightweight processing that operates directly on in-the-wild mixtures. The method combines an online Itakura–Saito nonnegative matrix factorization (NMF) update with an online auxiliary-function independent vector analysis (IVA) framework, preserving the low-rank spectral model employed in ILRMA while updating the demixing matrix frame by frame with bounded latency and memory. Simulations on multitrack music mixtures show improved separation accuracy and a real-time factor below one, indicating feasibility for live and interactive scenarios. These results suggest blind separation suitable for low-latency music applications.} }
Endnote
%0 Conference Paper %T Online Independent Low-Rank Matrix Analysis as a Lightweight and Trainable Model for Real-Time Multichannel Music Source Separation %A Taishi Nakashima %A Nobutaka Ono %B Proceedings of the AAAI 2026 Workshop on Audio-Centric AI: Towards Real-World Multimodal Reasoning and Application Use Cases (Audio-AAAI) %C Proceedings of Machine Learning Research %D 2026 %E Tatsuya Komatsu %E Keisuke Imoto %E Xiaoxue Gao %E Nobutaka Ono %E Nancy F. Chen %F pmlr-v312-nakashima26a %I PMLR %P 48--60 %U https://proceedings.mlr.press/v312/nakashima26a.html %V 312 %X In this paper, we propose an online extension of independent low-rank matrix analysis (ILRMA) for blind music source separation under real-time constraints. Because multitrack stems are rarely released, we target lightweight processing that operates directly on in-the-wild mixtures. The method combines an online Itakura–Saito nonnegative matrix factorization (NMF) update with an online auxiliary-function independent vector analysis (IVA) framework, preserving the low-rank spectral model employed in ILRMA while updating the demixing matrix frame by frame with bounded latency and memory. Simulations on multitrack music mixtures show improved separation accuracy and a real-time factor below one, indicating feasibility for live and interactive scenarios. These results suggest blind separation suitable for low-latency music applications.
APA
Nakashima, T. & Ono, N.. (2026). Online Independent Low-Rank Matrix Analysis as a Lightweight and Trainable Model for Real-Time Multichannel Music Source Separation. Proceedings of the AAAI 2026 Workshop on Audio-Centric AI: Towards Real-World Multimodal Reasoning and Application Use Cases (Audio-AAAI), in Proceedings of Machine Learning Research 312:48-60 Available from https://proceedings.mlr.press/v312/nakashima26a.html.

Related Material