Adaptive Echo Suppression Masks for Small-Speech Double-Talk Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing echo suppression devices fail to detect speech accurately when the speech is small and suppress echoes appropriately, leading to potential voice loss of the near-end speaker.
Innovation Solution
An echo suppression device that includes a mask storage unit, a mask selection unit, and a double-talk detection unit, which generates and selects optimal masks based on reception signal magnitude, allowing for accurate echo suppression even when speech is small.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed mask is generated assuming large signal in receiving signal path, then echo suppression is effective for large signals, but speech detection fails when speech is small and reception signal is large
Solution Approach 1:
The patent applies dynamics by making the mask adaptive rather than fixed. The mask is dynamically adjusted based on the actual magnitude of the reception signal through a mask selection unit that chooses from multiple pre-generated masks corresponding to different signal levels. This resolves the contradiction by allowing the system to switch between different mask characteristics depending on the current operating condition, ensuring both effective echo suppression for large signals and accurate speech detection for small signals.
Solution Approach 2:
The patent changes the parameter of mask magnitude to adapt to different reception signal levels. Multiple masks are generated with different magnitudes corresponding to different reception signal levels, and the appropriate mask is selected based on the current signal conditions. This parameter change enables the system to maintain optimal performance across varying signal conditions, resolving the contradiction between echo suppression effectiveness and speech detection accuracy.
2Object-generated harmful factors
If echo suppressor strongly acts on input signal for large reception signal, then echo is suppressed, but near-end speaker voice disappears
Solution Approach 1:
The patent changes the mask magnitude parameter based on reception signal level to control the strength of echo suppression. For large reception signals, a mask with appropriate magnitude is selected to suppress echo while preserving near-end speech. For small speech signals, a different mask is selected that provides gentler suppression, preventing voice disappearance. This dynamic parameter adjustment resolves the contradiction between echo suppression and information preservation.
Solution Approach 2:
The system dynamically adjusts the echo suppression strength by selecting from multiple masks with different characteristics. The mask selection unit chooses the appropriate mask based on the current reception signal magnitude, making the echo suppression adaptive rather than fixed. This dynamic adjustment ensures that echo is effectively suppressed when signals are large while preserving near-end speaker voice when signals are small.
3Device complexity
If single mask is used for all signal levels, then device complexity is low, but adaptability to varying reception signal levels is poor
Solution Approach 1:
The patent applies preliminary action by pre-generating multiple masks with different magnitudes corresponding to different reception signal levels before actual operation. These pre-generated masks are stored in a mask storage unit, and during operation, the appropriate pre-prepared mask is selected based on the current signal conditions. This approach provides good adaptability without requiring complex real-time mask generation, resolving the contradiction between device complexity and adaptability.
Data Source
AI summary
Even when a speech is small, the speech is allowed to be detected and an echo is allowed to be appropriately suppressed. Whenever a sample point of a reception signal transmitted through a receiving signal path that transmits a signal to a speaker is acquired, an optimal mask is sequentially generated or selected from base masks as one or a plurality of masks generated based on a learning signal based on a reception signal acquired within a predetermined period before a time point at which the sample point was acquired. Whenever the optimal mask is selected, whether a double-talk state is present is sequentially detected based on a result of comparing an input signal with the optimal mask. When detecting that a speech is not input to a microphone and the reception signal includes a speech, a process of suppressing an echo is sequentially performed on the input signal.