Hearing Device ML Noise Cancellation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users of hearing instruments experience poor speech quality and low speech intelligibility in challenging acoustical environments due to the presence of strong background noise, despite the use of beamforming technologies.
Innovation Solution
A hearing device equipped with an input module, a time domain filter, a processor, a receiver, and a controller featuring a machine learning model that processes input signals to determine a gain and filter control signal, thereby reducing background noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If beamforming technologies are used to suppress interfering sources, then speech intelligibility is improved, but strong background noise remains present in the desired direction
Solution Approach 1:
A machine learning model is introduced as an intermediary component between the beamforming stage and the final output. This ML model analyzes the spectral characteristics of the signal and generates gain values that selectively attenuate background noise while preserving speech content in the desired direction, thereby resolving the contradiction between speech intelligibility and background noise suppression
Solution Approach 2:
The system dynamically changes the gain parameter across different frequency bins based on ML-predicted speech presence probabilities. By adapting the gain parameter according to the spectral content and noise characteristics, the system achieves both improved speech intelligibility and effective background noise reduction without compromising signals from the desired direction
2Measurement precision
If machine learning model is added for noise cancellation, then speech quality is improved, but device complexity increases
Solution Approach 1:
The machine learning model is designed to perform multiple functions: it predicts speech presence, estimates noise levels, and generates gain values for noise suppression. By consolidating these functions into a single ML model, the system achieves improved speech quality without proportionally increasing device complexity, as one component handles multiple tasks that would otherwise require separate processing modules
Solution Approach 2:
The ML model is trained offline using labeled data and deployed as a pre-trained model in the hearing device. During operation, the model self-activates to process incoming signals without requiring real-time retraining or complex adaptive algorithms, thereby maintaining relatively simple device architecture while achieving sophisticated noise cancellation and speech quality enhancement
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A hearing device is disclosed, the hearing device comprising an input module for provision of an input signal. The input module comprises one or more microphones including a first microphone for provision of a first microphone input signal. The input signal is based on the first microphone input signal. The hearing device comprises a time domain filter for filtering the input signal for provision of a filter output signal. The hearing device comprises a processor for processing the filter output signal and providing an electrical output signal based on the filter output signal. The hearing device comprises a receiver for converting the electrical output signal to an audio output signal. The hearing device comprises a controller comprising a machine learning, ML, model for provision of an ML output based on the input signal. The controller is configured to determine a first gain based on the ML output. The controller is configured to determine a filter control signal based on the first gain. The controller is configured to provide the filter control signal to the time domain filter for filtering the input signal based on the filter control signal.