Residual Echo Suppression via Frequency Band Attenuation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Electronic devices face challenges in accurately isolating local speech from residual echo signals due to internal coupling and nonlinearity in the acoustic path, which degrades signal quality and interferes with wakeword detection during voice command processing.
Innovation Solution
The system performs acoustic echo cancellation and selectively applies residual echo suppression based on energy statistics, attenuating specific frequency bands with low signal quality metrics to improve wakeword detection by rectifying portions of the audio signal corresponding to the expected energy values of a keyword.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If acoustic echo cancellation is performed to remove echo signals, then echo suppression is improved, but residual echo and signal degradation still occur due to internal coupling and nonlinearity
Solution Approach 1:
The patent segments the audio signal processing into multiple stages: initial acoustic echo cancellation, followed by residual echo suppression, and then wakeword detection. By dividing the processing into discrete frequency bands and applying different suppression levels to each band based on echo presence, the system achieves more reliable signal quality while maintaining effective echo suppression.
Solution Approach 2:
The patent applies local quality by treating different frequency bands differently. Instead of uniform echo suppression across all frequencies, the system analyzes each frequency band separately and applies residual echo suppression only where needed. This localized approach maintains signal quality in clean frequency bands while suppressing echo in contaminated bands.
2Object-affected harmful factors
If residual echo suppression is applied to all frequency bands, then echo suppression is improved, but local speech components may be distorted or removed
Solution Approach 1:
The patent makes different parts of the audio signal (frequency bands) have different functions. Some bands receive aggressive echo suppression while others receive minimal or no suppression, depending on the detected echo presence and speech content in each band. This preserves speech fidelity in clean bands while removing echo in contaminated bands.
Solution Approach 2:
Instead of applying full echo suppression across all frequency bands, the patent applies partial suppression only to the extent necessary. The system selectively suppresses residual echo in specific frequency bands where it is present, avoiding over-suppression that would distort speech components in other bands.
3Object-affected harmful factors
If the entire audio signal is processed for echo suppression, then echo removal is improved, but processing time and computational resources increase
Solution Approach 1:
The patent segments the audio signal into multiple frequency bands and processes each band independently. This allows parallel processing of different bands, reducing overall processing time compared to sequential full-signal processing. The segmented approach also enables selective processing of only those bands containing echo.
Solution Approach 2:
The patent applies echo suppression partially rather than universally. By identifying and processing only the frequency bands that contain residual echo, the system reduces computational workload and processing time compared to applying suppression to the entire audio signal regardless of echo presence.
4Object-affected harmful factors
If acoustic echo cancellation is performed, then echo signals are reduced, but distortion and signal degradation occur due to imprecise time alignment
Solution Approach 1:
The patent segments the echo cancellation process into frequency-band-specific operations. By analyzing and correcting time alignment issues in each frequency band separately, the system can achieve more accurate alignment than global time correction, compensating for variations in the acoustic path across different frequencies.
Solution Approach 2:
The patent applies different time alignment corrections to different frequency bands based on their specific characteristics. This localized correction approach addresses the nonlinearity and frequency-dependent behavior of the acoustic path, improving alignment accuracy compared to uniform time correction applied to all frequencies.
Data Source
AI summary
A system configured to improve wakeword detection. The system may selectively rectify (e.g., attenuate) a portion of an audio signal based on energy statistics corresponding to a keyword (e.g., wakeword). For example, a device may perform echo cancellation to generate isolated audio data, may use the energy statistics to calculate signal quality metric values for a plurality of frequency bands of the isolated audio data, and may select a fixed number of frequency bands (e.g., 5-10%) associated with lowest signal quality metric values. To detect a specific keyword, the system determines a threshold λ(f) corresponding to an expected energy value at each frequency band. During runtime, the device determines signal quality metric values by subtracting residual music from the expected energy values. Thus, the device attenuates only a portion of the total number of frequency bands that include more energy than expected based on the energy statistics of the wakeword.