Hearing Aid Neural Network Tuning for User-Specific Loss Emulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hearing aids equipped with fixed neural networks perform suboptimally due to their inability to adapt to individual users and varying acoustic situations, leading to ineffective noise reduction and hearing loss compensation across a wide range of sound input levels and frequencies.
Innovation Solution
A method is introduced that optimizes the training parameters of neural networks in hearing aids by determining frequency and level distributions based on equalized sound pressure levels, allowing the network to better emulate both normal-hearing and hearing-impaired auditory models, thereby improving performance across different acoustic conditions without increasing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a fixed neural network is used in all hearing aids, then device complexity is reduced and ease of manufacture is improved, but adaptability to individual users and acoustic situations deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training neural networks offline for different acoustic situations and hearing loss configurations before deployment. Multiple specialized neural networks are prepared in advance, each optimized for specific conditions, and then selected during operation based on the current acoustic environment and user profile, eliminating the need for real-time adaptive training while maintaining high adaptability.
Solution Approach 2:
The patent implements dynamics by making the neural network selection dynamic based on real-time acoustic situation detection and user profile matching. The system dynamically switches between different pre-trained neural networks according to the current operational context, providing adaptability without requiring the neural network structure itself to change during operation.
2Adaptability or versatility
If a neural network is trained to span a large dynamic input range, then coverage of sound input levels is improved, but performance across different sound levels and frequencies deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the large dynamic input range into multiple smaller, specialized training ranges. Each neural network is trained on a specific segment of the acoustic environment (e.g., quiet, moderate, noisy conditions) and specific frequency ranges, allowing each network to achieve high precision within its designated segment while the system as a whole covers the full dynamic range through selective deployment.
Solution Approach 2:
The patent implements local quality by optimizing each neural network for specific local conditions rather than attempting a single universal solution. Each network is tailored to perform exceptionally well in its specific acoustic situation and frequency range, with different networks having different optimization characteristics suited to their local operational context.
3Adaptability or versatility
If different neural networks are used for different users and acoustic situations, then adaptability and performance are improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent applies preliminary action by performing all complex neural network training and adaptation work offline before the hearing aid is deployed. Multiple user-specific and situation-specific neural networks are pre-computed and stored in the device, eliminating the need for complex real-time training capabilities while maintaining high adaptability through selective network deployment.
Solution Approach 2:
The patent implements copying by creating multiple copies of neural networks with different training configurations for different users and acoustic situations. Rather than implementing a single complex adaptive system, the device stores multiple pre-trained network copies and selects the appropriate one based on the current operational context, simplifying the real-time processing requirements.
Data Source
AI summary
A method of defining and setting a signal processing of a hearing aid is disclosed. The hearing aid is configured to be worn by a user at or in an ear of the user. The method comprises providing at least one electric input signal representing at least one input sound signal from a sound environment of a hearing aid user, determining a normal-hearing representation of said at least one electric input signal based on a selected normal-hearing auditory model fj, determining optimised training parameters of a neural network, where the neural network represents a hearing-impaired representation of said at least one electric input signal based on a hearing-impaired auditory model, wherein determining the optimised training parameters comprises determining a frequency distribution, βj, and a level and frequency distribution, αj,1, of said at least one electric input signal based on an equalization of sound pressure levels of said at least one electric input signal. A hearing aid adapted to be worn in or at an ear of a user is furthermore disclosed.