Joint Neural Network Hearing Aid Audio for Feedback and Noise

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing hearing aid technologies struggle to simultaneously optimize speech enhancement and acoustic feedback cancellation due to the inability of current machine-learning methods to simulate acoustic feedback and account for interactions between modules, leading to artifacts like chirping and howling.

Innovation Solution

A deep neural network is trained to simultaneously calibrate speech enhancement and feedback cancellation by simulating the acoustic feedback path, using a closed-loop system to optimize parameters and reduce background noise and artifacts.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If separate machine-learning models are used for speech enhancement and feedback cancellation, then each module can be optimized independently, but interactions between modules cause artifacts like chirping and howling

Engineering Contradiction:
ImproveIndependent module optimizationVSAvoidArtifact-free output
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent combines separate speech enhancement and feedback cancellation models into a single joint neural network model. This unified model processes audio signals through multiple pathways simultaneously - one pathway handles speech enhancement while another handles feedback cancellation, allowing both functions to operate cooperatively without generating artifacts. The joint training approach optimizes all parameters together, ensuring consistent behavior across different operating conditions.

Inventive Principle:
Principle #5Merging (Combining)

2Ease of manufacture

If current machine-learning methods are used without acoustic feedback simulation, then training is simpler, but the system cannot account for acoustic feedback interactions

Engineering Contradiction:
ImproveTraining simplicityVSAvoidAcoustic feedback handling
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent implements a simulated acoustic feedback path within the training framework. The neural network output is fed back through a simulated acoustic model that replicates real-world feedback conditions. This allows the model to learn how to handle acoustic feedback during training, improving its adaptability to real-world scenarios while maintaining training feasibility through efficient simulation techniques.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent introduces a simulated acoustic feedback model as an intermediary component during training. This virtual feedback path acts as a mediator between the neural network output and the training data, allowing the system to learn feedback cancellation without requiring complex real-world feedback scenarios. The simulation bridge enables transfer of learned skills to actual acoustic feedback conditions.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If a unified model is used for joint optimization, then module interactions are handled correctly, but the device complexity increases

Engineering Contradiction:
ImproveOptimized system performanceVSAvoidNeural network architecture
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The unified neural network is segmented into distinct functional pathways that handle different tasks. One pathway processes speech enhancement while another handles feedback cancellation, with each pathway having specialized sub-layers. This segmentation maintains the benefits of joint optimization while making the overall architecture more manageable and interpretable, reducing effective complexity through structured organization.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent designs a multi-functional neural network that performs both speech enhancement and feedback cancellation within a single unified architecture. The network uses shared lower layers for common signal processing tasks and diverges into specialized pathways for different functions. This universal approach consolidates multiple models into one, reducing overall device complexity while maintaining optimized performance for both tasks.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4243449B1Apparatus and method for speech enhancement and feedback cancellation using a neural network
Publication Date: 2026.04.15 STARKEY LABORATORIES INC
  • EP4243449B1 patent drawingFigure 1
  • EP4243449B1 patent drawingFigure 2~3
  • EP4243449B1 patent drawingFigure 4

AI summary

A hearing device includes a deep/recurrent neural network trained to jointly perform sound enhancement and feedback cancellation. During training a neural network is connected between a simulated input and a simulated output of the hearing device. The neural network is operable to change a response affecting the simulated output. The neural network is trained by applying the simulated input to the deep neural network while applying the feedback path response between the simulated input and the simulated output. The deep-neural network is trained to reduce an error between the simulated output and the reference audio signal and used for sound enhancement in the device.