Hearing Aid Neural Network for Speech Enhancement and Feedback Cancellation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hearing aid technologies struggle to simultaneously optimize speech enhancement and feedback cancellation due to the inability of current machine-learning methods to account for acoustic feedback and its interaction with other modules, leading to artifacts like chirping and instability.
Innovation Solution
A deep neural network is trained to jointly optimize speech enhancement and feedback cancellation by simulating the acoustic feedback path and adjusting the feedback canceller step-size, using a closed-loop simulation to enhance sound quality and reduce artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate machine-learning methods are used for speech enhancement and feedback cancellation, then each function can be optimized independently, but the system produces artifacts like chirping and instability due to inability to account for acoustic feedback interactions
Solution Approach 1:
The patent merges speech enhancement and feedback cancellation into a single joint machine-learning model. The model processes the microphone signal through shared layers that simultaneously perform both functions, eliminating the need for separate processing chains and their associated instability issues.
Solution Approach 2:
The patent incorporates an explicit feedback path in the neural network architecture that models the acoustic feedback loop. This feedback mechanism allows the model to learn and compensate for acoustic feedback interactions, preventing chirping artifacts and instability.
2Reliability
If a unified deep neural network is used to jointly optimize speech enhancement and feedback cancellation, then system stability and sound quality improve, but the model complexity and training requirements increase
Solution Approach 1:
The unified neural network is segmented into distinct functional layers: speech enhancement layers, feedback cancellation layers, and intermediate processing layers. This segmentation allows the complex model to be trained incrementally and deployed efficiently on hearing aid hardware.
Solution Approach 2:
The patent designs a universal neural network architecture that performs multiple functions (speech enhancement, feedback cancellation, and acoustic modeling) within a single model, reducing overall system complexity compared to maintaining separate specialized models for each function.
3Measurement precision
If closed-loop simulation is used during training to account for acoustic feedback, then the model achieves better real-world performance, but the training process becomes more computationally intensive
Solution Approach 1:
The patent performs closed-loop simulation and acoustic feedback modeling during the offline training phase, allowing the model to learn realistic feedback characteristics before deployment. This preliminary action ensures high training accuracy while minimizing real-time computational energy consumption during actual hearing aid operation.
Data Source
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.