Global Rotation Equivariant Phase Modeling for Speech Enhancement with Deep Magnitude-Phase Interaction

2026-02-09

AI Summary

This work addresses the limitation of conventional deep learning approaches that model speech phase in Euclidean space, failing to capture its intrinsic circular topology and thereby constraining phase reconstruction performance. To overcome this, the authors propose a manifold-aware magnitude-phase dual-stream framework, which introduces global rotation equivariance (GRE) into phase modeling for speech enhancement—a first in the field. They design two novel components: a Magnitude-Phase Interactive Convolutional Module (MPICM) and a Hybrid-Attention Dual-FFN (HADF) module, both preserving GRE properties to enable deep fusion between magnitude and phase representations. Experimental results demonstrate that the proposed method reduces phase distance by over 20% in phase reconstruction tasks, improves zero-shot cross-corpus denoising PESQ scores by more than 0.1, and achieves superior performance in general-purpose speech enhancement under mixed distortions.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Mixture of Experts (MoE)
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

While deep learning has advanced speech enhancement (SE), effective phase modeling remains challenging, as conventional networks typically operate within a flat Euclidean feature space, which is not easy to model the underlying circular topology of the phase. To address this, we propose a manifold-aware magnitude-phase dual-stream framework that aligns the phase stream with its intrinsic circular geometry by enforcing Global Rotation Equivariance (GRE) characteristic. Specifically, we introduce a Magnitude-Phase Interactive Convolutional Module (MPICM) for modulus-based information exchange and a Hybrid-Attention Dual-FFN (HADF) bottleneck for unified feature fusion, both of which are designed to preserve GRE in the phase stream. Comprehensive evaluations are conducted across phase retrieval, denoising, dereverberation, and bandwidth extension tasks to validate the superiority of the proposed method over multiple advanced baselines. Notably, the proposed architecture reduces Phase Distance by over 20\% in the phase retrieval task and improves PESQ by more than 0.1 in zero-shot cross-corpus denoising evaluations. The overall superiority is also established in universal SE tasks involving mixed distortions. Qualitative analysis further reveals that the learned phase features exhibit distinct periodic patterns, which are consistent with the intrinsic circular nature of the phase. The source code is available at https://github.com/wangchengzhong/RENet.

Problem

• phase modeling
• speech enhancement
• circular topology
• rotation equivariance
• magnitude-phase interaction

Innovation

• Global Rotation Equivariance
• Phase Modeling
• Manifold-aware Framework
• Magnitude-Phase Interaction
• Circular Topology