Neural Network Hearing Aid Signal Path for Speech-in-Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional hearing aids struggle to effectively separate speech from background noise in noisy environments due to limitations in computational power and the impracticality of incorporating neural networks, leading to decreased speech intelligibility for individuals with hearing loss.
Innovation Solution
A dual-path signal processing system in hearing aids that integrates a neural network engine (NNE) and a digital signal processor (DSP), allowing selective engagement of neural network-based audio enhancement, with a controller determining the processing path based on user inputs, environmental factors, and sensor data to optimize user experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural network algorithms are incorporated into hearing aids to separate speech from background noise, then speech intelligibility in noisy environments is improved, but power consumption and computational requirements increase beyond what hearing aid batteries can provide
Solution Approach 1:
The system segments the neural network processing into two parts: a training phase that occurs offline on powerful computers to generate processing rules, and an inference phase that occurs on the hearing aid using pre-processed acoustic features. This segmentation allows complex speech separation without requiring the hearing aid to perform computationally intensive neural network training, thus reducing power consumption while maintaining speech intelligibility improvement.
Solution Approach 2:
The system performs preliminary processing of acoustic signals by extracting relevant features (such as spectral characteristics, temporal patterns) before feeding them to the neural network. This preliminary feature extraction reduces the dimensionality and complexity of the input data, enabling the neural network to operate more efficiently on the hearing aid with reduced computational power and lower energy consumption while still achieving effective speech separation.
2Measurement precision
If traditional signal processing techniques are used in hearing aids, then power consumption remains low, but the ability to separate speech from background noise is insufficient
Solution Approach 1:
The system introduces an intermediary layer between the raw acoustic signal and the neural network processing. This intermediary layer consists of hand-crafted acoustic feature extractors that transform the raw signal into a simplified representation containing essential speech and noise characteristics. This intermediary representation enables the neural network to focus on higher-level speech separation tasks without being overwhelmed by the full complexity of raw audio data, thus improving noise separation capability while managing device complexity.
3Measurement precision
If complex neural network models are deployed in hearing aids, then speech separation performance improves, but latency increases which affects real-time processing
Solution Approach 1:
The system extracts and removes the computationally intensive training function from the hearing aid device, placing it in an external training environment. The hearing aid retains only the lightweight inference function that applies pre-learned processing rules to incoming audio. This extraction of the training component allows the use of sophisticated neural network models for speech separation while maintaining low latency in real-time processing, as the inference stage requires minimal computation.
Data Source
AI summary
The disclosure generally relates to a method, system and apparatus for processing audio through a neural network contained in a hearing device. In one embodiment, the disclosure relates to an apparatus to enhance incoming audio signal. The apparatus includes a controller to receive an incoming signal and provide a controller output signal; neural network engine (NNE) circuitry in communication with the controller, the NNE circuitry activatable by the controller, the NNE circuitry configured to generate an NNE output signal from the controller output signal; and digital signal processing (DSP) circuitry to receive one or more of controller output signal or the NNE circuitry output signal to thereby generate a processed signal; wherein the controller determines a processing path of the controller output signal through one of the DSP or the NNE circuitries as a function of one or more of predefined parameters, incoming signal characteristics and NNE circuitry feedback.