Multi-Microphone Voice Audio Processing for Wind Noise Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Wind noise detected by microphones interferes with voice audio recordings, causing unpleasant and unnatural artifacts in video recordings.
Innovation Solution
An image capture device with multiple microphones determines coherence values across frequency bands to identify and separate wind noise from voice audio, selecting non-voice sub-band frequency bins with the lowest energy and voice sub-band frequency bins based on thresholds or coherence values to generate a composite signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple microphones are used to capture audio, then the ability to identify and separate wind noise from voice audio is improved, but the device complexity increases
Solution Approach 1:
The audio spectrum is segmented into different frequency bins, allowing the system to analyze and process wind noise and voice audio separately in different frequency ranges. This segmentation enables precise identification of wind noise characteristics without requiring complex hardware modifications.
Solution Approach 2:
Coherence values are introduced as an intermediary metric to compare signals from multiple microphones. By calculating coherence across frequency bins, the system can objectively identify wind noise without directly analyzing the complex multi-microphone signal relationships, simplifying the processing approach.
2Measurement precision
If coherence values are calculated across all frequency bins, then the precision of wind noise identification is improved, but the processing time and computational energy increase
Solution Approach 1:
The system applies different processing strategies to different frequency bins based on their characteristics. Voice sub-bands and non-voice sub-bands are treated differently, with selection criteria tailored to each band's properties. This local optimization reduces overall processing requirements while maintaining accuracy.
Solution Approach 2:
The system changes parameters dynamically by selecting frequency bins based on energy values and coherence thresholds. By adapting the selection criteria to the actual signal characteristics rather than processing all bins uniformly, the system achieves precise wind noise identification with reduced computational load.
3Object-affected harmful factors
If non-voice sub-band frequency bins are selected based on lowest energy values, then wind noise removal is improved, but the risk of removing valid low-energy voice components increases
Solution Approach 1:
The system uses coherence values as feedback to verify selections made based on energy criteria. By cross-referencing coherence measurements with energy-based selections, the system can confirm whether low-energy bins truly represent wind noise or valid voice components, reducing the risk of erroneous removal.
Solution Approach 2:
The system performs preliminary analysis by calculating coherence values across all frequency bins before making selection decisions. This preliminary step establishes a baseline for identifying wind noise characteristics, allowing subsequent selection processes to be more accurate and less likely to remove valid voice components.
Data Source
AI summary
An image capture device determines a coherence value between two or more microphones. The microphone signals produced by the two or more microphones each include a non-voice sub-band and a voice sub-band. The non-voice sub-band and the voice sub-band each comprise frequency bins. The coherence value is measured per bin for each of the microphone signals. The non-voice sub-band frequency bins from the first microphone signal and the second microphone signal that have the lowest energy value are selected for generating a composite signal. The voice sub-band frequency bins from a predetermined microphone signal are selected for generating the composite signal. Alternatively, the voice sub-band bins can be selected based on the average minimum energy across the voice band. The composite signal that includes the selected non-voice sub-band frequency bins and the voice sub-band frequency bins is output to a memory of the image capture device.