Hearing Implant Audio Scene Classification With Two-Stage Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cochlear implant systems face challenges in accurately classifying audio scenes due to the nearly infinite variations and seamless transitions, leading to inefficient training of neural networks and potential misclassification, which affects the quality of sound perception.
Innovation Solution
A two-step neural network training process is employed, separating the network into a pre-processing neural network and a classification neural network, allowing for optimized training with smaller datasets and faster convergence using meta-parameters optimization techniques like CMA-ES and MBO, ensuring robust audio scene classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single neural network is used for audio scene classification, then the network can process the entire classification task, but the training requires extensive data and computational resources leading to slow convergence
Solution Approach 1:
The patent divides the single neural network into two separate networks: a pre-processing neural network that extracts features from audio signals, and a classification neural network that performs the actual scene classification. This segmentation allows each network to be specialized and trained more efficiently on specific tasks, reducing overall training time and computational requirements while maintaining classification accuracy.
2Measurement precision
If extensive training data is used to improve classification accuracy, then the model generalization improves, but the training time and computational resources increase significantly
Solution Approach 1:
The pre-processing neural network extracts essential features and meta-parameters from the audio signals before passing them to the classification network. This extraction process condenses the information content, allowing the classification network to achieve high accuracy with less training data since it operates on extracted features rather than raw audio signals, thereby reducing training time.
Solution Approach 2:
The pre-processing neural network performs preliminary feature extraction and optimization before the classification stage. By preparing the data in advance through this preliminary action, the classification network receives pre-processed input that requires less training to achieve accurate classification, thus reducing the overall training time and computational resources needed.
3Measurement precision
If the neural network is optimized for specific audio scenes, then the classification accuracy for those scenes improves, but the ability to generalize to new audio scenes decreases
Solution Approach 1:
The pre-processing neural network is designed to extract universal features and meta-parameters that are applicable across different audio scenes. This universal pre-processing stage enables the classification network to adapt to new audio scenes more effectively, improving generalization capability while maintaining accurate classification for known scenes through the specialized classification layer.
Data Source
AI summary
An audio scene classifier classifies an audio input signal from an audio scene and includes a pre-processing neural network configured for pre-processing the audio input signal based on initial classification parameters to produce an initial signal classification, and a scene classifier neural network configured for processing the initial scene classification based on scene classification parameters to produce an audio scene classification output. The initial classification parameters reflect neural network training based on a first set of initial audio training data, and the scene classification parameters reflect neural network training on a second set of classification audio training data separate and different from the first set of initial audio training data. A hearing implant signal processor configured for processing the audio input signal and the audio scene classification output to generate the stimulation signals to the hearing implant for perception by the patient as sound.