Hearing Device Audio Processing Customization via Acoustic Scene Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hearing devices lack the ability to effectively customize audio signal processing for individual users' specific acoustic environments, leading to suboptimal hearing experiences and reliability issues.
Innovation Solution
A method utilizing an acoustic scene classification unit with an embedding neural network and classifiers to classify and customize audio signal processing by adding personalized acoustic scene classes, allowing for automatic recognition of environments without user intervention, using a two-stage classification process and few-shot training techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If hearing devices use fixed audio signal processing settings, then device complexity is reduced, but adaptability to different acoustic environments deteriorates
Solution Approach 1:
The hearing device dynamically adapts its audio signal processing by automatically detecting acoustic scenes and switching between different processing settings. The system transitions from static fixed settings to dynamic environment-responsive processing, improving adaptability while managing complexity through automated decision-making algorithms.
Solution Approach 2:
The hearing device performs self-adjustment by automatically detecting the acoustic environment and selecting appropriate processing parameters without user intervention. This self-service capability enhances adaptability while reducing the need for complex manual configuration, effectively managing the complexity-adaptability trade-off.
2Adaptability or versatility
If hearing devices require manual user configuration for personalized settings, then customization capability is improved, but ease of operation deteriorates
Solution Approach 1:
The system automatically detects acoustic scenes and applies appropriate processing settings without requiring user configuration. This self-service approach maintains strong customization capability through environment-aware processing while dramatically improving ease of operation by eliminating manual setup requirements.
Solution Approach 2:
The hearing device continuously monitors the acoustic environment and automatically adjusts processing parameters based on real-time feedback from scene detection. This closed-loop feedback mechanism enables effective customization while maintaining simple operation, as the system self-adjusts without user input.
3Measurement precision
If hearing devices use extensive training data for accurate classification, then measurement precision is improved, but loss of time deteriorates
Solution Approach 1:
The system performs preliminary training offline using extensive training data to establish accurate acoustic scene classification models. Once trained, the model can quickly classify new environments without requiring extensive real-time training, thus achieving high measurement precision while minimizing operational time loss through pre-computed classification rules.
Solution Approach 2:
The training process is segmented into an offline phase where extensive data is processed to create classification models, and an online phase where the pre-trained models rapidly classify acoustic scenes. This segmentation allows high precision through comprehensive training data while avoiding time loss during actual operation, as the heavy computational work is completed beforehand.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A hearing device (2) and a method for customizing audio signal processing of the hearing device (2) are described. The hearing device (2) is configured for audio signal processing of an input audio signal (I) to obtain an output audio signal (O), using an acoustic scene classification unit (12) for classifying the input audio signal (I) to belong to one or more acoustic scene classes from a set of acoustic scene classes. For customizing the audio signal processing, at least one audio signal sample of an acoustic environment is provided to an embedding neural network (14) of the acoustic scene classification unit (12). A sample embedding for each audio signal sample is determined using the embedding neural network (14). The audio signal processing is customized by retraining at least one classifier (15) of the acoustic scene classification unit (12) using the sample embeddings, thereby adding an acoustic scene class for the acoustic environment to the set of acoustic scene classes.