Sub-band Mixing for Consistent Timbre in Multi-Mic Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio mixing technologies using multiple microphones face challenges with noticeable fluctuations in timbre and sound level due to gating, and increased noise and reverberation when multiple talkers are active simultaneously.
Innovation Solution
The implementation of subband mixing techniques that transform microphone signals into frequency audio data, apply weights based on smoothed spectral powers and noise floors, and integrate signals across multiple microphones to enhance direct speech and suppress noise and reverberations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If gating is used to mute microphones when active talkers are not detected, then microphone signal processing is simplified, but fluctuations in perceived timbre and sound level become noticeable
Solution Approach 1:
The patent divides the audio frequency spectrum into multiple subbands and processes each subband separately. This segmentation allows the system to apply different gain values to different frequency regions, smoothing transitions when microphones are switched and maintaining consistent perceived timbre while still enabling selective microphone muting based on talker detection.
Solution Approach 2:
The patent applies different processing characteristics to different subbands rather than uniformly across the entire frequency spectrum. By calculating subband-specific gain values based on local energy distribution and talker presence, the system maintains optimal signal quality in each frequency region while reducing audible artifacts during microphone transitions.
2Object-affected harmful factors
If multiple microphones are processed by gating, then noise and reverberation can be reduced, but when multiple talkers are active simultaneously, noise and reverberation levels in the audio mix increase
Solution Approach 1:
The patent segments the audio spectrum into multiple subbands and independently processes each subband. This allows the system to identify and suppress noise and reverberation in each frequency region separately, while still preserving multiple simultaneous talkers by maintaining their respective subband energy distributions. The segmented approach prevents the accumulation of noise and reverberation that occurs with traditional full-band gating methods.
3Device complexity
If traditional audio mixing is used with multiple microphones, then implementation is simpler, but speech intelligibility and direct sound enhancement are reduced
Solution Approach 1:
The patent implements subband processing that divides the audio spectrum into multiple frequency regions, allowing selective enhancement of direct speech components in each subband. This segmentation enables the system to amplify speech-related frequency components while suppressing noise and reverberation, significantly improving speech intelligibility compared to traditional full-band mixing approaches.
Data Source
AI summary
Input audio data portions of a common time window index value generated by multiple microphones at a location are received. Subband portions are generated from the input audio data portions. Peak powers, noise floors, etc., are individually determined for the subband portions. Weights for the subband portions are computed based on the peak powers, the noise floors, etc., for the subband portions. An integrated audio data portion of the common time window index is generated based on the subband portions and the weight values for the subband portions. An integrated signal may be generated based at least in part on the integrated audio data portion.