Hearing Aid Algorithm Training Using Masked In-Domain Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training algorithms to extract desired components from sound signals, such as those used in hearing aids, face challenges due to mismatches between training data and real-world sound signals, leading to imprecise target extraction.
Innovation Solution
A method for training an algorithm that uses partly masked in-domain data elements to determine the parameters of an encoder, optimizing the prediction of noisy components and improving the accuracy of target extraction while maintaining a reasonable algorithm size suitable for small battery-driven devices like hearing aids.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If simulated sound signals are used for training, then training data availability is improved, but mismatch between training data and real-world signals increases leading to imprecise target extraction
Solution Approach 1:
The patent introduces an acoustic model as an intermediary component that bridges the gap between simulated training data and real-world audio signals. The acoustic model processes the simulated noisy speech signals to better align their characteristics with actual hearing aid input, thereby reducing the domain mismatch while maintaining training data availability
Solution Approach 2:
The patent modifies parameters of the training data by applying acoustic models that adjust the statistical and spectral characteristics of simulated signals. This transforms the training data parameters to better match real-world conditions, improving target extraction precision without sacrificing data availability
2Measurement precision
If algorithm complexity is increased to improve target extraction accuracy, then extraction precision is improved, but device resource requirements increase making it unsuitable for battery-driven devices
Solution Approach 1:
The patent extracts and focuses on training only the essential encoder component of the neural network architecture, rather than training the entire complex system. This selective extraction of the core functional element achieves good target extraction accuracy while significantly reducing the overall algorithm size and resource requirements for deployment in battery-driven devices
Data Source
AI summary
Disclosed herein are embodiments of a method for training an algorithm. Training can include an encoder and a first decoder with partly masked in-domain data element(s). Further, the method can include determining a parameter for optimizing predictions. Further disclosed herein are embodiments of hearing aids utilizing one or more of the trained algorithms.