Cochlear Encoder-Decoder Network for Real-Time Hearing Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cochlear models for hearing aids and machine-hearing applications are computationally expensive due to their cascaded nature, making them unsuitable for real-time processing, and they fail to capture inner-hair-cell and auditory-nerve processing accurately.
Innovation Solution
A multilayer convolutional encoder-decoder neural network is used to emulate cochlear processing, incorporating nonlinear units and shortcut connections to mimic cochlear filter tuning, allowing for fast and accurate emulation of basilar membrane vibrations and frequency selectivity, suitable for parallel processing on GPUs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional signal processing methods are used in hearing aids, then device complexity is reduced, but the ability to accurately model cochlear mechanics and provide effective hearing assistance for severe-to-profound hearing loss is insufficient
Solution Approach 1:
The patent replaces traditional mechanical signal processing methods with a neural network-based computational model. The neural network learns cochlear mechanics from training data and provides hearing assistance through adaptive signal processing, achieving superior performance for severe-to-profound hearing loss while maintaining computational efficiency through pre-trained models.
2Measurement precision
If a neural network model with high accuracy is used to model cochlear mechanics, then measurement precision of cochlear response is improved, but computational resources and processing time increase
Solution Approach 1:
The neural network model is pre-trained offline using extensive training data to learn cochlear mechanics. This preliminary action allows the model to achieve high measurement precision when deployed in hearing aids, as the complex computational work has already been performed during the training phase rather than during real-time signal processing.
3Reliability
If hearing aids are optimized for severe-to-profound hearing loss, then effectiveness for the target population is improved, but adaptability to other hearing loss types is reduced
Solution Approach 1:
The neural network model uses adjustable parameters including hearing loss type, degree, and configuration as inputs. By changing these parameters, the same model architecture can adapt to different hearing loss types (sensorineural, conductive, mixed) and degrees (mild to profound), providing optimized hearing assistance for each condition while maintaining the ability to handle severe-to-profound cases.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A method and hearing device (100) for emulating cochlear processing of auditory stimuli are disclosed, in which a multilayer convolutional encoder-decoder neural network (10) sequentially compresses and then decompresses a time-domain input comprising a plurality of samples. At least one nonlinear unit for applying a nonlinear transformation is mimicking a level-dependent cochlear filter tuning associated with cochlear mechanics and outer hair cells. Other described modules cover inner-hair-cell and auditory-nerve fiber processing. A plurality of shortcut connections (15) is directly forwarding inputs between convolutional layers of the encoder (11) and the decoder (12). An output layer (14) is generating, for each input to the neural network, N output sequences of cochlear response parameters corresponding to N emulated cochlear filters associated with N different center frequencies to span a cochlear tonotopic place-frequency map. A transducer (105) of the hearing device converts output sequences generated by the neural network (10) into auditory-stimulus dependent audible time-varying pressure signals, or basilar-membrane vibrations, inner-hair-cell potentials, auditory-nerve firing patterns or population coding thereof for auditory or augmented hearing applications.