Residual Echo Suppression Using Delayed Spectral Mismatch Recalculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio systems, particularly in vehicles, face challenges in completely removing residual echo signals after double-talk conditions end, leading to errors in transfer function mismatch estimates, which can result in either over or under-suppression of speech signals.
Innovation Solution
Implementing a method that resumes spectral mismatch calculation after a delay period following the end of a double-talk period and adjusting time smoothing constants based on the current talk status, allowing for accurate spectral mismatch estimates by distinguishing between single-talk and double-talk periods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If spectral mismatch calculation is resumed immediately after double-talk period ends, then calculation speed is improved, but calculation accuracy deteriorates due to speech component corruption
Solution Approach 1:
The system performs preliminary action by implementing a delay period after double-talk detection before resuming spectral mismatch calculation. This delay allows the speech component to decay sufficiently, ensuring that when calculation resumes, the microphone signal contains minimal speech contamination. The delay period is determined based on the decay characteristics of speech signals, balancing calculation speed with accuracy.
2Speed
If time smoothing constant is set to smooth out speech components quickly, then transition response is improved, but spectral mismatch accuracy during single-talk period deteriorates
Solution Approach 1:
The system applies dynamic adjustment of the time smoothing constant based on the operational phase. During the transition phase from double-talk to single-talk, a smaller smoothing constant is used to quickly smooth out speech components. Once the transition to single-talk is confirmed and the delay period has elapsed, a larger smoothing constant is applied to achieve accurate spectral mismatch estimation during steady-state single-talk operation.
3Object-generated harmful factors
If echo suppression is applied using corrupted spectral mismatch values, then echo suppression is improved, but speech signal quality deteriorates due to over or under suppression
Solution Approach 1:
The system employs feedback mechanisms by continuously monitoring the operational state (double-talk or single-talk conditions) and using this information to control when spectral mismatch calculation should be paused or resumed. This feedback loop ensures that spectral mismatch values are only updated when the microphone signal is free from speech components, preventing corruption and ensuring reliable speech signal quality in the suppressed output.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of suppressing residual echo includes calculating a spectral mismatch of an acoustic echo canceler based upon a program content signal and a residual signal; determining a set of filter coefficients based at least in part upon a selected spectral mismatch; filtering the residual signal based upon the set of filter coefficients; freezing the calculation of the selected spectral mismatch in response to detecting a double talk condition in the residual signal; waiting a predetermined hold period in response to detecting that the double talk condition has ended; and, after the predetermined hold period, resuming the calculation of the spectral mismatch based upon the program content signal and the residual signal.