Wind Noise Mitigation via Frequency Bin Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio devices face challenges in distinguishing user voice commands from wind noise, leading to reduced intelligibility and false activations in noisy environments.
Innovation Solution
A wind noise mitigation system using a multi-stage comparison masking method and optimal microphone blending, which samples sound signals from multiple microphones to differentiate between correlated speech and uncorrelated wind noise, attenuating the latter to improve signal-to-noise ratio and preserve speech intelligibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If wind noise is present in the audio environment, then the audio device can still capture sound signals, but the intelligibility of voice control inputs is reduced and false activations occur
Solution Approach 1:
The audio signal is segmented into multiple frequency bins through Fourier transformation, allowing individual processing of different frequency components. This segmentation enables selective attenuation of noise-containing frequency bins while preserving speech-containing bins, thereby improving voice control accuracy in windy conditions
Solution Approach 2:
A reference microphone is introduced as an intermediary element that captures primarily wind noise without the user's voice. This reference signal serves as a mediator to identify and subtract wind noise components from the main microphone signals, enhancing the reliability of voice control by removing the harmful wind noise interference
2Measurement precision
If multiple microphones are used to capture sound signals, then the ability to differentiate speech from wind noise is improved, but the device complexity increases
Solution Approach 1:
The wind noise component is extracted from the composite audio signal by comparing multiple microphone inputs. The system identifies and separates the wind noise portion from the speech portion based on correlation analysis, then removes the extracted wind noise to improve measurement precision while managing processing complexity through efficient algorithms
3Object-affected harmful factors
If wind noise is attenuated aggressively, then speech intelligibility is improved, but desired speech signals may also be attenuated causing loss of information
Solution Approach 1:
Different attenuation levels are applied to different frequency bins based on their individual noise and speech characteristics. Frequency bins dominated by wind noise receive aggressive attenuation, while bins containing speech signals receive minimal or no attenuation. This local quality approach ensures wind noise is reduced without losing speech information
Solution Approach 2:
The system continuously analyzes the audio signal to identify wind noise components and adjusts attenuation levels in real-time. By providing feedback on the detected noise levels and speech content, the system dynamically optimizes the attenuation process to reduce wind noise while preserving speech integrity, preventing information loss
Data Source
AI summary
A system and method can provide nose, such as wind noise, mitigation and/or microphone blending. Some methods may include sampling a sound signal from a plurality of microphones to generate a frame comprising a plurality of time-frequency tiles of the sound signal, each time-frequency tile including respective values of at least one feature from the plurality of microphones, comparing the respective values of the at least one feature to determine whether each time-frequency tile satisfies a similarity threshold, and flagging each time-frequency tile as noise if it fails to satisfy the similarity threshold, grouping the plurality of time-frequency tiles into sets of frequency-adjacent time-frequency tiles, and for each set of frequency-adjacent time-frequency tiles in the frame: counting a number of flagged time-frequency tiles, and attenuating all of the time-frequency tiles in the each set if the number exceeds a noise bin count threshold to thereby reduce noise in the sound signal.