Speech Enhancement Mask Adjustment for Noise Pumping Control

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Mask-based speech enhancement algorithms suffer from noise pumping/gating issues, particularly in low-SNR conditions, leading to inconsistencies between dialog and non-dialog segments in the denoised audio output, which can annoy listeners.

Innovation Solution

A noise compensation method that estimates residual noise using a convex function of the mask and audio signal magnitude, followed by determining a modified mask to stabilize noise suppression across segments, reducing audible inconsistencies.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If mask-based algorithms remove all noise within non-dialog segments, then noise suppression performance is improved, but inconsistencies between dialog and non-dialog segments are generated causing audible pumping/gating effects

Engineering Contradiction:
Improvenoise suppression performanceVSAvoidaudible pumping/gating effects
Core Design Contradiction:
ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent applies parameter changes by modifying the mask values through a convex function transformation. The original binary mask (0 or 1) is transformed using a convex function to create intermediate values, which smooths the transition between dialog and non-dialog segments. This resolves the contradiction by changing the parameter representation of the mask to eliminate abrupt changes that cause pumping/gating effects while maintaining noise suppression effectiveness.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If mask-based algorithms filter noise in low-SNR conditions, then noise removal capability is improved, but inconsistencies between segments become more pronounced

Engineering Contradiction:
Improvenoise removal capabilityVSAvoidconsistency between segments
Core Design Contradiction:
ReliabilityVSStability of the object's composition

Solution Approach 1:

The convex function transformation changes the parameter values of the mask to create a more stable output. By applying the convex function to the mask values, the patent ensures that even in low-SNR conditions where aggressive noise removal is needed, the transformed mask provides smoother transitions and more consistent behavior across different segment types, thereby improving stability while maintaining noise removal capability.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4430606B1Improving noise compensation in mask-based speech enhancement
Publication Date: 2025.11.05 DOLBY LABORATORIES LICENSING CORP
  • EP4430606B1 patent drawingFigure 1
  • EP4430606B1 patent drawingFigure 2
  • EP4430606B1 patent drawingFigure 3

AI summary

Methods and apparatus for improving noise compensation in mask-based speech enhancement are described. A method of processing an audio signal, which includes one or more speech segments, includes obtaining a mask for mask-based speech enhancement of the audio signal and obtaining a magnitude of the audio signal. An estimate of residual noise is determined in the audio signal after mask-based speech enhancement, based on the mask and the magnitude of the audio signal. A modified mask is determined based on the estimate of the residual noise. Further described are corresponding programs and computer-readable storage media.