Echo Removal via Dual-Model Accuracy Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing echo cancellation and suppression techniques face challenges in accurately estimating and removing echo signals, particularly when both echo and near-end signals are strong, leading to divergence and poor model estimates, and require complex doubletalk detection and step-size control algorithms.
Innovation Solution
The method employs two models, where a first model is always updated and a second model is updated only when the first model is deemed more accurate, avoiding the need for doubletalk detection and step-size control, and using these models to estimate and suppress echo signals in received audio signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional echo cancellation techniques are used to estimate and remove echo signals, then echo suppression is achieved, but model divergence occurs and accuracy deteriorates when both echo and near-end signals are strong
Solution Approach 1:
The patent divides the single echo model into two separate models: a first model that continuously tracks echo parameters and a second model that provides stable echo estimates. This segmentation allows each model to specialize - the first model adapts to changing conditions while the second model maintains stability, resolving the contradiction between tracking accuracy and estimation reliability when both echo and near-end signals are present.
Solution Approach 2:
The patent introduces a selective updating mechanism that acts as an intermediary between the two models. Instead of directly updating the echo model during double-talk conditions, the system uses the first model to track changes and selectively updates the second model only when accuracy is confirmed. This intermediary process prevents divergence while maintaining accurate echo estimates.
2Measurement precision
If doubletalk detection and step-size control algorithms are implemented to improve echo cancellation, then adaptation accuracy is enhanced, but system complexity increases
Solution Approach 1:
The patent extracts and eliminates the need for complex doubletalk detection and step-size control algorithms by using a simpler architecture with two models where the first model continuously updates and selectively informs the second model. This extraction of unnecessary complexity components maintains adaptation accuracy while significantly reducing algorithmic complexity compared to traditional approaches.
3Speed
If echo model is continuously updated to track changing echo conditions, then adaptation speed increases, but model stability deteriorates leading to divergence
Solution Approach 1:
The patent implements dynamic model updating where the first model continuously adapts to track changing echo conditions at high speed, while the second model remains stable by being updated only selectively based on accuracy comparisons. This dynamic approach allows the system to maintain both fast adaptation capability and model stability, preventing divergence while tracking echo variations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Echo removal techniques are described. As part of the echo removal, a first model estimate of the echo in the received audio signal is determined using a first model and a second model estimate of the echo is determined using a second model. A first accuracy value of the first model is determined according to a model accuracy measure, and a second accuracy value of the second model is determined according to the model accuracy measure. It is then determined if the first model is more accurate than the second model based on a comparison of the first accuracy value and the second accuracy value and the second model is selectively updated based on said comparison. Echo removal is applied to the received audio signal using only the second model estimate of the echo.