Audio Noise Reduction via Spatial Feature Vector Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional dual microphone noise reduction methods for headsets with widely separated microphones, such as in-ear headsets, face challenges in dynamic noise environments due to limited noise correlation and spatial aliasing, leading to insufficient noise reduction, especially under windy conditions.
Innovation Solution
A noise reduction method that derives spatial features from cross-correlation and energy calculations between microphones, uses adaptive spatial filtering, and estimates power spectral density to differentiate near-field speech from background noise, with dynamic threshold adjustments to handle varying noise levels and wind noise effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If microphones are widely separated to improve spatial diversity for noise reduction, then noise reduction capability is improved, but spatial aliasing occurs and noise correlation is limited
Solution Approach 1:
The system dynamically adapts to changing array orientations and noise conditions by continuously tracking the user's speech direction and adjusting processing parameters. This allows the widely separated microphones to maintain effective noise reduction despite dynamic changes in spatial relationships and environmental conditions.
Solution Approach 2:
The system changes processing parameters based on detected conditions, including adaptive spatial filtering coefficients and dynamic threshold adjustments for wind noise detection. This enables the system to optimize performance across varying noise levels and wind conditions while using the fixed wide microphone separation.
2Measurement precision
If adaptive spatial filtering is used to reduce background noise, then noise reduction is improved, but performance degrades under windy conditions due to high energy transient signals
Solution Approach 1:
The system converts the harmful effect of wind noise into a detectable signal characteristic. By analyzing the high energy transient signals caused by wind, the system identifies wind noise presence and activates specialized processing modes that suppress these transient signals while preserving speech content.
Solution Approach 2:
The system introduces an intermediary detection mechanism that identifies wind noise conditions before they severely degrade speech quality. This intermediary detection layer allows the system to switch processing strategies specifically for wind conditions, separating wind noise handling from general background noise reduction.
3Measurement precision
If microphones are placed on ear bud to improve spatial separation, then spatial diversity is improved, but physical dimension constraints prevent optimal placement
Solution Approach 1:
The system transitions from relying solely on physical microphone separation within the ear bud to utilizing directional signal processing in the spatial domain. By processing signals to enhance speech from the user's mouth direction and suppress noise from other directions, the system achieves effective spatial separation without requiring additional physical space in the ear bud.
Data Source
AI summary
An audio device has an array of microphones and a voice processing system that obtains a multi-dimensional spatial feature vector comprising at least a correlation of the microphones and a calculation of at least one ratio of energies of the microphones, uses the multi-dimensional feature vector to estimate an energy of near-field speech and background noise, uses a ratio of the near-field speech energy and background noise estimates to estimate a probability of a presence of the near-field speech, adaptively combines signals from the microphones based on the estimated near-field speech presence probability to provide a combined output signal comprising a near-field speech signal and a residual background noise signal, estimates a power spectral density of the residual background noise signal present at the combined output signal using the estimated near-field speech presence probability, and reduces the background noise by using the estimated power spectral density.