An Investigation on Combining Geometry and Consistency Constraints into Phase Estimation for Speech Enhancement

TL;DR

This paper introduces a new iterative phase estimation method called MSGLA for speech enhancement, combining geometry and consistency constraints to improve noise suppression.

Contribution

It presents a novel multi-source Griffin-Lim algorithm that integrates consistency and geometric constraints for more effective phase reconstruction in speech enhancement.

Findings

MSGLA variants match or outperform existing algorithms.

Effective noise suppression demonstrated on benchmark datasets.

Validation through oracle experiments confirms effectiveness under ideal conditions.

Abstract

We propose a novel iterative phase estimation framework, termed multi-source Griffin-Lim algorithm (MSGLA), for speech enhancement (SE) under additive noise conditions. The core idea is to leverage the ad-hoc consistency constraint of complex-valued short-time Fourier transform (STFT) spectrograms to address the sign ambiguity challenge commonly encountered in geometry-based phase estimation. Furthermore, we introduce a variant of the geometric constraint framework based on the law of sines and cosines, formulating a new phase reconstruction algorithm using noise phase estimates. We first validate the proposed technique through a series of oracle experiments, demonstrating its effectiveness under ideal conditions. We then evaluate its performance on the VB-DMD and WSJ0-CHiME3 data sets, and show that the proposed MSGLA variants match well or slightly outperform existing algorithms, including direct phase estimation and DNN-based sign prediction, especially in terms of background noise suppression.