Noise-Level Speech Mixing in Ear-Worn Devices to Reduce Pumping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise reduction technologies in ear-worn devices struggle to effectively mix speech and noise based on noise level, leading to potential pumping effects when relying on signal-to-noise ratio (SNR).
Innovation Solution
The solution involves using a neural network to predict speech and noise components of an audio signal, and then mixing these components based on the level of the noise component, rather than relying solely on SNR. This approach uses functions that relate the weight applied to the noise component with the noise level, potentially resulting in U-shaped or Z-shaped weight functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If noise reduction is performed using SNR-based mixing, then speech and noise can be separated and mixed, but pumping effects occur due to dynamic fluctuations in the mixing ratio
Solution Approach 1:
The patent changes the parameter used for controlling the mixing ratio from SNR to noise level. Specifically, it uses the level of the noise component (derived from minimum statistics or neural network prediction) as the control parameter instead of signal-to-noise ratio. This parameter substitution eliminates the pumping effects while maintaining effective noise reduction, as the noise level provides a more stable basis for mixing ratio adjustment.
Solution Approach 2:
The patent replaces the traditional SNR-based control mechanism with a noise-level-based control mechanism. By substituting the control logic from SNR-dependent mixing to noise-level-dependent mixing, the system achieves smoother dynamic behavior without the harmful pumping artifacts that characterize SNR-based approaches.
2Reliability
If noise component is completely removed, then noise reduction is maximized, but distortion increases and environmental awareness is lost
Solution Approach 1:
Instead of completely removing the noise component, the patent applies partial action by mixing back a controlled portion of the noise component into the speech signal. The mixing ratio is dynamically adjusted based on noise level, allowing complete noise removal when appropriate while preserving some noise for naturalness and environmental awareness when needed, thus avoiding distortion.
Solution Approach 2:
The patent implements dynamic mixing where the proportion of noise component in the output signal varies continuously based on the current noise level. This dynamic adjustment allows the system to adapt between complete noise removal and noise preservation modes, optimizing both noise reduction performance and signal naturalness according to environmental conditions.
3Reliability
If noise component is completely removed, then noise reduction is maximized, but environmental awareness for the wearer is reduced
Solution Approach 1:
The patent applies partial noise removal by dynamically mixing back appropriate amounts of noise component based on environmental noise level. This preserves essential environmental cues for wearer awareness while still achieving effective noise reduction, avoiding complete elimination of the noise component which would loss environmental information.
Solution Approach 2:
The system uses feedback from noise level estimation to control the mixing ratio. The estimated noise level continuously informs the mixing decision, creating a closed-loop system that maintains environmental awareness by preserving noise components when environmental awareness is needed while achieving noise reduction when appropriate.
Data Source
AI summary
An ear-worn device may include neural network circuitry configured to determine, using a neural network, one or both of a neural network-predicted speech component of an input audio signal and a neural network-predicted noise component of the input audio signal. The ear-worn device may further include mixing circuitry configured to perform mixing resulting in an output signal corresponding to the neural network-predicted speech component of the input audio signal mixed with the neural network-predicted noise component of the input audio signal. The ear-worn device may further include adaptive mixing control circuitry configured to control the mixing performed by the mixing circuitry based, at least in part, on a level of an estimate of a stationary noise component of the input audio signal and/or a level of the neural network-predicted noise component of the input audio signal.