Hearing Aid Noise Reduction via Power-Constrained Spectral Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
State-of-the-art multi-microphone speech enhancement systems face challenges in accurately estimating speech and noise spectra, often resulting in negative spectral components and overestimation of speech power due to noise influences, which affects the quality of noise reduction in hearing aids.
Innovation Solution
A power-constraint enforced maximum likelihood estimator is used to improve speech quality by considering inter-frequency bin relationships, ensuring non-negative and unbiased estimates of speech power spectral densities across frequency bins, thereby enhancing noise reduction without compromising speech intelligibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If existing methods estimate speech and noise spectra independently for each frequency bin, then the problem is simplified, but inter-bin relationships between spectral components are precluded
Solution Approach 1:
The frequency spectrum is divided into multiple frequency bins, and the estimation problem is segmented into independent per-bin estimations using maximum likelihood estimators. This segmentation simplifies the overall problem while maintaining accuracy through subsequent power constraint enforcement that reconstructs inter-bin relationships.
Solution Approach 2:
The solution moves from one-dimensional per-bin estimation to a multi-dimensional approach by enforcing power constraints across the frequency dimension. The power constraint Σkλs(k,l) = Ps(l) connects all frequency bins through a shared power relationship, enabling inter-bin relationships to be utilized while maintaining computational tractability.
2Ease of operation
If maximum likelihood estimation is applied without power constraints, then estimation is straightforward, but negative spectral components and overestimation of speech power occur
Solution Approach 1:
The power constraint Σkλs(k,l) = Ps(l) is applied as a preliminary condition to prevent negative spectral components and overestimation. By enforcing this constraint before final estimation, the method proactively eliminates validity issues rather than correcting them afterward, ensuring reliable non-negative speech power spectral density estimates.
Solution Approach 2:
The estimation process changes from unconstrained maximum likelihood estimation to constrained estimation by introducing the power parameter relationship. This parameter change transforms the estimation problem to enforce physical reality (non-negative power) while maintaining the simplicity of maximum likelihood approaches through Lagrange multiplier methods.
3Productivity
If per-bin independent estimation is used, then computational simplicity is achieved, but speech quality and noise reduction performance are compromised
Solution Approach 1:
The solution merges per-bin estimation results through the power constraint relationship. By combining individual bin estimates under the unified constraint Σkλs(k,l) = Ps(l), the method achieves both computational efficiency of per-bin processing and speech enhancement quality of integrated spectral estimation, resolving the contradiction between productivity and precision.
Data Source
AI summary
A hearing aid comprises a) a multitude of M input transducers each providing an electric input signal representative of environment sound in a time-frequency representation (k, l), and each comprise varying amounts of target (s) and noise (v) signal components; b) a signal processor configured to process said multitude of electric input signals; and comprising a beamformer filter configured to receive said multitude M of electric input signals and to provide a spatially filtered signal and a post-filter configured to receive said spatially filtered signal and to provide an estimate Ŝ(k,l) of a target signal representing said target signal components from said target sound source. The signal processor is configured to provide estimates of power spectral densities λs(k,l) of said target signal components in dependence of inter-frequency bin relationships between the spectral components enforced by properties of the electric input signals across at least some of said frequency bins.