Spatial Noise Suppression Gain Selection for Single-Mic Audio

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing noise suppression algorithms for voice communications on edge devices struggle with background noise transmission, leading to degraded voice intelligibility due to low signal-to-noise ratios, especially on devices with a single microphone and in varying noise environments.

Innovation Solution

A machine learning assisted approach using a neural network for spatial noise estimation and suppression, combining directionality and level models to estimate noise and speech probabilities, and apply noise suppression gains based on microphone covariance and power spectra, without requiring multiple microphones.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional noise suppression algorithms are used on edge devices, then device cost is reduced, but noise suppression performance deteriorates due to low signal-to-noise ratio

Engineering Contradiction:

Improvenoise suppression performanceVSAvoidbackground noise

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent introduces an intermediary statistical model (Gaussian distribution model) that bridges the input audio signal and the noise suppression output. The model estimates speech and noise probabilities, means, and covariances to create an intermediate representation that enables effective noise suppression even with single-microphone input, resolving the contradiction between device simplicity and noise suppression performance

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms the audio signal processing by changing key parameters: it estimates and utilizes speech probability, noise probability, speech mean, noise mean, speech covariance, and noise covariance for each frequency band. These parameter transformations enable the system to distinguish speech from noise and apply targeted suppression, improving performance without requiring multiple microphones

Inventive Principle:

Principle #35Parameter changes

2Reliability

If multiple microphones are used to improve noise suppression, then signal-to-noise ratio increases, but device complexity and cost increase

Engineering Contradiction:

Improvesignal-to-noise ratioVSAvoidmicrophone configuration

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent creates a statistical copy or model of the multi-microphone noise suppression effect using single-microphone input. By estimating speech and noise covariances and applying probabilistic models, the system replicates the noise suppression performance that would traditionally require multiple microphones, thereby maintaining high signal-to-noise ratio while avoiding the complexity of multiple microphone hardware

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces the mechanical solution of using multiple microphones with a computational/statistical approach. Instead of relying on physical microphone arrays to spatially filter noise, the system uses machine learning models and statistical estimations to achieve noise suppression, substituting mechanical complexity with algorithmic processing

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If complex noise suppression processing is applied, then noise suppression performance improves, but computational load and processing time increase

Engineering Contradiction:

Improvenoise suppression qualityVSAvoidreal-time processing capability

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent segments the audio signal processing into distinct frequency bands and processes each band independently using separate statistical models. This segmentation allows parallel processing of different frequency components, reducing the computational burden on each processing unit while maintaining overall noise suppression quality, thus enabling real-time processing on edge devices

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial processing by focusing computational resources only on frequency bands where noise is actually present, rather than processing the entire spectrum uniformly. By estimating noise presence in each band and applying suppression only where needed, the system reduces unnecessary computational load while maintaining effective noise suppression quality

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12555591B2Machine learning assisted spatial noise estimation and suppression

Publication Date: 2026.02.17 DOLBY LABORATORIES LICENSING CORP

AI summary

In an embodiment, a method comprises: receiving bands of power spectra of an input audio signal and a microphone covariance, and for each band: estimating, using a classifier, respective probabilities of speech and noise; estimating, using a directionality model, a set of means for speech and noise, or a set of means and covariances for speech and noise, based on the microphone covariance for the band and the probabilities; estimating, using a level model, a mean and covariance of noise power based on the probabilities and the power spectra; determining a first noise suppression gain based on the directionality model; determining a second noise suppression gain based on the level model; selecting the first or second noise suppression gain or their sum based on a signal-to-noise ratio of the input audio signal; and scaling a time-frequency representation of the input signal by the selected noise suppression gain.