Adaptive Filter Divergence Detection for Hands-Free Echo Cancellation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional echo cancellation systems in hands-free devices struggle to effectively distinguish between double talk situations and echo-only situations, leading to inadequate echo cancellation and degradation of useful speech during double talk periods, due to complex algorithms and inefficient detection methods.
Innovation Solution
A method that detects double talk by observing behaviors characteristic of adaptive filter divergence, deriving a representative index, and using this index to control the adaptation step of the echo cancellation algorithm, allowing continued adaptation during double talk periods with reduced speed and maintaining efficient echo cancellation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional double talk detection algorithms are used to detect double talk situations, then detection capability is provided, but the algorithms are relatively complex and demand high computing power without providing high certainty
Solution Approach 1:
The patent extracts the essential characteristic of double talk situations by focusing solely on the convergence status of the adaptive filter, rather than analyzing the full spectral envelope. This extraction simplifies the detection algorithm while maintaining detection capability, directly resolving the contradiction between detection accuracy and algorithm complexity
Solution Approach 2:
The adaptive filter's own convergence behavior serves as the detection criterion. The system uses the filter's intrinsic performance metric (convergence status) to detect double talk situations, eliminating the need for separate complex detection algorithms and reducing computational demands while maintaining detection accuracy
2Object-generated harmful factors
If echo cancellation post-processing with variable gain is applied to attenuate residual echo, then echo reduction is achieved, but useful speech is simultaneously degraded during double talk periods
Solution Approach 1:
The patent dynamically adjusts the adaptation step of the echo cancellation algorithm based on the detected double talk situation. During double talk periods, the adaptation step is reduced to prevent speech degradation, while normal operation uses standard adaptation. This dynamic adjustment resolves the contradiction by adapting the cancellation strength to the current operational context
Solution Approach 2:
The system changes the adaptation step parameter based on the convergence status indicator. When double talk is detected (filter has not converged), the adaptation step is modified to maintain speech quality. This parameter change allows the system to reduce residual echo during normal operation while preserving useful speech during double talk periods
3Stability of the object's composition
If the adaptive filter is frozen during double talk to prevent divergence, then echo cancellation stability is maintained, but the filter cannot track slow variations of the acoustic path
Solution Approach 1:
The patent implements dynamic adaptation step adjustment based on the detected double talk situation. During double talk periods, the adaptation step is reduced rather than completely frozen, allowing the filter to maintain stability while still tracking slow acoustic path variations. This dynamic approach resolves the contradiction between stability and adaptability
Solution Approach 2:
The system changes the adaptation parameter (step size) based on the operational context. During double talk, a smaller adaptation step is used to maintain stability while preserving tracking capability. This parameter modification allows the filter to remain both stable and adaptable, resolving the contradiction between these two requirements
Data Source
Figure 1~2
AI summary
The method involves echo cancellation processing, using echo cancellation module (40) by implementing adaptive linear filter algorithm. Double talk situation is detected, by evaluating index representative of degree of convergence divergence of algorithm, and assessing whether predetermined condition is satisfied. The double talk situation arisen on basis of assessment result is deduced. One parameter of adaptive linear filter algorithm is modified in event of double talk arising in response to detecting double talk situation.