Hearing Microphone DOA Estimation Using Neural Network Phase Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hearing devices struggle to accurately determine the direction of arrival (DOA) of sound sources, which affects their ability to enhance speech intelligibility and adapt to challenging listening environments.
Innovation Solution
The implementation of a classifier neural network that processes audio signals from multiple microphones to estimate the DOA by transforming time domain signals into frequency transform streams, extracting phase information, and classifying features to predict the angles of arrival, utilizing machine learning models like CNNs or RNNs for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional signal processing methods are used for DOA estimation, then device complexity is reduced, but measurement precision of direction of arrival is insufficient
Solution Approach 1:
The patent replaces traditional mechanical signal processing methods with machine learning-based neural networks to estimate direction of arrival. The neural network model processes frequency transform streams and extracts directional features, achieving superior accuracy compared to conventional approaches while managing computational complexity through efficient model architecture.
Solution Approach 2:
The patent transforms time domain signals into frequency transform streams and extracts specific features from these transformed signals. By changing the parameter space from time domain to frequency domain and selecting appropriate feature representations, the system achieves more accurate DOA estimation while optimizing processing complexity.
2Measurement precision
If machine learning models are implemented for DOA estimation, then measurement precision improves, but use of energy increases
Solution Approach 1:
The patent applies partial action by processing only the necessary portions of the audio signal - specifically extracting features from frequency transform streams at key frequency buckets rather than processing the entire signal spectrum. This selective processing reduces energy consumption while maintaining sufficient accuracy for DOA estimation.
Solution Approach 2:
The patent extracts only the essential features from the frequency transform streams that are necessary for DOA estimation. By taking out and processing only these critical features rather than the complete signal data, the system achieves accurate directional information with reduced computational and energy requirements.
3Measurement precision
If multiple microphones are used to improve DOA estimation, then measurement precision improves, but device complexity increases
Solution Approach 1:
The patent implements multi-functionality by using the same microphone array for multiple purposes: capturing acoustic signals, estimating direction of arrival, and enabling adaptive sound processing. The microphone array serves both as a simple audio capture device and as a directional sensing system, maximizing utility while managing complexity.
Data Source
AI summary
A hearing device has two or more different microphones and a processor operably coupled to the two or more microphones. The processor is operable via instructions to obtain time domain signals from the microphones and transform the time domain signals into respective frequency transform streams. The frequency transform streams characterize phase information in the time domain signals. The processor further determines, from the frequency transform streams, features indicative of directional information of one or more sound sources. The processor inputs the features to a classifier neural network. The classifier network is trained to classify the features into probabilities of angles of arrival of the sound sources. Based on the classification of the features, the processor estimates one or more directions of arrival (DOA) of the sound sources.