Dynamic RTF Estimation via Structured Sparse Bayesian Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing multi-microphone speech processing systems in hearing devices fail to adapt to environmental changes, such as head movements and external noise interference, leading to inaccurate speech intelligibility and quality due to ineffective noise reduction methods.

Innovation Solution

A dynamic Relative Transfer Function (RTF) estimation using Structured Sparse Bayesian Learning (S-SBL) is implemented, which incorporates a hierarchical Bayesian framework for unified treatment of sparse early reflections and exponential decaying reverberation, providing a robust and efficient method for improving speech processing by estimating the RTF within a short burst of noisy recordings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If traditional Time Domain least square approach is used for RTF estimation, then the method is simple to implement, but the estimates become ineffective and unstable due to noise and finite samples

Engineering Contradiction:
ImproveImplementation simplicityVSAvoidEstimation stability
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent transforms the RTF estimation problem from time domain to frequency domain, changing the mathematical representation parameters. This allows the use of spectral subtraction and logarithmic operations that are more robust to noise and finite sample effects, thereby improving estimation stability while maintaining computational feasibility

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces the direct time-domain least squares mechanical approach with a frequency-domain statistical approach using spectral analysis and Bayesian inference. This substitution introduces probabilistic modeling that inherently handles noise and uncertainty, improving reliability without significantly increasing implementation complexity

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If existing beamformers with simple geometric assumptions are used, then the system is computationally efficient, but the assumptions do not adapt to movement, external noise interference, or other changes in acoustic environment

Engineering Contradiction:
ImproveComputational efficiencyVSAvoidAdaptability to environmental changes
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic RTF estimation that adapts to changing acoustic environments by continuously updating the relative transfer function based on observed audio signals. This dynamic approach replaces static geometric assumptions with adaptive statistical modeling that responds to head movements, noise interference, and environmental changes while maintaining computational efficiency through iterative optimization

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent incorporates feedback mechanisms where the estimated RTF is used to improve subsequent estimations. The system uses the computed RTF to enhance speech separation and noise reduction performance, which in turn provides better training data for refining the RTF estimates, creating a self-improving adaptive system that responds to environmental changes

Inventive Principle:
Principle #23Feedback

3Measurement precision

If dynamic RTF estimation is implemented to adapt to environmental changes, then speech intelligibility and quality improve, but the complexity of the processing system increases

Engineering Contradiction:
ImproveSpeech intelligibilityVSAvoidProcessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex RTF estimation problem into manageable frequency bins and time frames, processing each independently through spectral subtraction and logarithmic operations. This segmentation allows the use of simple per-bin calculations that accumulate to produce accurate overall estimates, improving speech intelligibility while keeping individual processing steps computationally simple

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by focusing computational resources only on the essential components of RTF estimation - the spectral magnitude relationships between microphones - rather than attempting to model all acoustic parameters. This selective approach achieves sufficient precision for speech enhancement without the complexity of complete acoustic scene analysis

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3148213B1Dynamic relative transfer function estimation using structured sparse bayesian learning
Publication Date: 2018.09.12 STARKEY LABORATORIES INC
  • EP3148213B1 patent drawingFigure 1
  • EP3148213B1 patent drawingFigure 2
  • EP3148213B1 patent drawingFigure 3

AI summary

The use of a dynamic Relative Transfer Function (RTF) between two or more microphones may be used to improve multi-microphone speech processing applications. The dynamic RTF may improve speech intelligibility and speech quality in the presence of environmental changes, such as variations in head or body movements, variations in hearing device characteristics or wearing positions, or variations in room or environment acoustics. The use of an efficient and fast dynamic RTF estimation algorithm using short burst of noisy, reverberant mic recordings, which will be robust to head movements may provide more accurate RTFs which may lead to a significant performance increase.