Mobile Speech Enhancement With Adaptive DNNs for Hearing Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hearing devices face challenges in effectively enhancing speech quality and intelligibility due to noise interference, particularly in complex environments, and require efficient sound processing techniques that can operate within practical power and processing limitations.
Innovation Solution
The use of simplified deep neural networks (DNNs) for sound enhancement in hearing devices, which are dynamically selected and adapted based on sound classification, combined with feature extraction and pruning techniques to reduce computational resources, allowing for efficient noise suppression and speech emphasis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for sound enhancement, then speech quality and intelligibility are improved, but power consumption and processing requirements increase
Solution Approach 1:
The patent applies parameter changes by dynamically selecting and adapting deep neural network models based on sound classification. The system adjusts the complexity and parameters of the DNN models according to the acoustic environment, using simpler models for straightforward conditions and more complex models only when necessary, thereby optimizing the balance between speech enhancement quality and power consumption.
Solution Approach 2:
The system implements dynamics by making the neural network selection adaptive rather than static. The audio signal is continuously classified, and the DNN model is dynamically switched based on the classification results. This dynamic adaptation allows the hearing device to respond to varying acoustic conditions in real-time, optimizing performance while managing power resources efficiently.
2Measurement precision
If deep neural networks are used for sound enhancement, then speech quality and intelligibility are improved, but processing requirements and device complexity increase
Solution Approach 1:
The patent segments the sound enhancement process into distinct stages: audio signal classification, DNN model selection, and speech enhancement processing. By dividing the system into these functional segments, the complexity is managed more effectively, with each segment performing a specific function and the overall system achieving enhanced speech quality through coordinated operation of simpler modular components.
Solution Approach 2:
The system dynamically adapts the processing requirements by selecting appropriate DNN models based on real-time sound classification. Rather than always using the most complex models, the system adjusts its processing intensity according to the acoustic environment, reducing computational burden in simple conditions while reserving complex processing for challenging noisy environments.
3Use of energy by moving object
If simplified deep neural networks are used, then power and processing requirements are reduced, but speech enhancement effectiveness may be compromised
Solution Approach 1:
The patent changes the parameters of the neural network models based on the classification of the audio signal. When the acoustic environment is simple, the system uses models with fewer parameters to conserve power. When the environment is complex and noisy, the system switches to models with more parameters to maintain speech enhancement effectiveness, thus optimizing the trade-off between power consumption and speech quality.
Solution Approach 2:
The system dynamically adjusts the complexity of the neural network processing based on real-time acoustic conditions. This dynamic approach ensures that simplified models are used whenever possible to reduce power consumption, while more sophisticated models are activated only when the acoustic environment demands enhanced processing capabilities to maintain speech intelligibility.
Data Source
AI summary
A system includes a mobile device that receives an audio signal from a microphone of the mobile device. The mobile device processes the audio signal via a neural network to obtain a speech-enhanced audio signal. The system includes an ear-wearable device comprising a data interface operable to communicate with the external data interface of the mobile device. The ear-wearable device includes an audio processing path coupled to the data interface and is operable to receive the speech-enhanced audio signal and reproduce the speech-enhanced audio in an ear of a user.