Spatial Filter Using Instantaneous Direction-of-Arrival Estimates

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing spatial filters struggle to adapt quickly to changing situations, such as source movements or competing speakers, due to their reliance on inaccurate signal models and limited ability to combine multiple microphone signals effectively, leading to degraded performance in noisy and reverberant conditions.

Innovation Solution

A spatial filter that generates audio output signals by using instantaneous direction-of-arrival information and noise statistics to compute complex weights for each time-frequency bin, allowing for adaptive noise reduction and dereverberation, and combining all available microphone signals to achieve an enhanced output.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If parametric spatial filters use a simple single plane wave signal model with instantaneous estimates, then the filter can quickly adapt to new situations, but the performance degrades when the signal model is violated in practice

Engineering Contradiction:

Improveadaptation speed to new situationsVSAvoidperformance under model violations

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent changes the signal model parameters from a simple single plane wave model to a more complex model that accounts for multiple sound sources and reverberation. By using multiple instantaneous direction-of-arrival estimates and combining them with noise statistics, the filter maintains adaptability while improving reliability under realistic conditions where signal models are often violated.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If linear spatial filters use accurate signal models and information, then optimal extraction is achieved, but the filters cannot adapt quickly to source movements or competing speakers

Engineering Contradiction:

Improveextraction optimalityVSAvoidadaptation to source movements

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent introduces dynamic adaptation by computing multiple instantaneous direction-of-arrival estimates for each time-frequency bin and using these to dynamically adjust the filter weights. This allows the filter to quickly adapt to source movements and competing speakers while maintaining optimal extraction performance through the use of accurate noise statistics and coherent combination of multiple estimates.

Inventive Principle:

Principle #15Dynamics

3Reliability

If all available microphone signals are used to estimate model parameters, then better noise reduction is achieved, but it is not straight-forward to combine multiple microphone signals to find an enhanced output signal

Engineering Contradiction:

Improvenoise reduction performanceVSAvoidcomplexity of combining multiple signals

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the processing by first estimating multiple direction-of-arrival parameters separately for each microphone signal in each time-frequency bin, then coherently combining these estimates using weighting based on noise statistics. This segmented approach simplifies the combination process while maintaining optimal noise reduction performance.

Inventive Principle:

Principle #1Segmentation

4Adaptability or versatility

If spatial filters use instantaneous direction-of-arrival estimates, then quick adaptation is achieved, but the estimates can be inaccurate in noisy and reverberant conditions

Engineering Contradiction:

Improvequick adaptation capabilityVSAvoiddirection-of-arrival estimation accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent merges multiple instantaneous direction-of-arrival estimates from different microphone signals by coherently combining them using weights derived from noise statistics. This combination process averages out estimation errors and improves the overall accuracy of the direction-of-arrival estimates while preserving the quick adaptation capability of instantaneous estimates.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS10331396B2Filter and method for informed spatial filtering using multiple instantaneous direction-of-arrival estimates

Publication Date: 2019.06.25 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV

AI summary

A filter for generating an audio output signal includes a plurality of audio output signal samples based on two or more input microphone signals. The audio output signal and the two or more input microphone signals are represented in a time-frequency domain, wherein each of the plurality of audio output signal samples is assigned to a time-frequency bin of a plurality of time-frequency bins. The filter includes a weights generator being adapted to receive, for each of the plurality of time-frequency bins, direction-of-arrival information and weighting information. Moreover, the filter includes an output signal generator for generating the audio output signal.