Cascaded Adaptive Interference Cancellation for Speech-Preserving Echo Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic echo cancellation techniques often inadvertently attenuate desired speech while attempting to isolate it from noise and echo signals, especially when the desired speech is obscured by noise, leading to inaccurate detection and processing.
Innovation Solution
A cascaded adaptive interference cancellation system that performs two stages of noise cancellation, where the first stage generates speech mask data indicating the presence of local speech, and the second stage adjusts the adaptive filter's adaptation based on this data to minimize attenuation of desired speech.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If acoustic echo cancellation techniques are used to isolate desired speech from noise and echo signals, then noise and echo are reduced, but desired speech is inadvertently attenuated
Solution Approach 1:
The patent divides the interference cancellation process into multiple stages: first stage performs interference cancellation on all signals, second stage performs speech detection on the processed signals, and third stage performs adaptive interference cancellation only when speech is not detected. This segmentation allows the system to cancel interference effectively while protecting desired speech through conditional processing.
Solution Approach 2:
The patent implements dynamic adaptation by adjusting the adaptive filter's behavior based on speech detection results. When speech is detected in the first output audio data, the system freezes adaptation of the adaptive filter to prevent attenuation of desired speech. When no speech is detected, the system resumes adaptation to continue canceling interference. This dynamic control resolves the contradiction between interference cancellation and speech preservation.
2Object-affected harmful factors
If adaptive filter continuously adapts to cancel interference, then interference cancellation is improved, but desired speech is distorted
Solution Approach 1:
The patent uses speech detection as a feedback mechanism to control the adaptive filter's adaptation process. The speech detection module analyzes the first output audio data and provides feedback to the adaptive interference cancellation module. When speech is detected, the feedback signal freezes adaptation; when no speech is detected, adaptation continues. This feedback loop ensures high signal fidelity by preventing the adaptive filter from distorting desired speech while maintaining effective interference cancellation.
Solution Approach 2:
The adaptive filter transitions between two operational states based on real-time conditions: adaptation mode when no speech is present, and frozen mode when speech is detected. This dynamic state change allows the system to optimize interference cancellation during non-speech periods while protecting speech fidelity during speech periods, resolving the contradiction between cancellation effectiveness and signal fidelity.
3Device complexity
If single stage noise cancellation is performed, then processing is simple, but speech detection accuracy is poor when speech is obscured by noise
Solution Approach 1:
The patent segments the noise cancellation process into two distinct stages: first stage performs interference cancellation to reduce noise and echo, generating first output audio data; second stage performs speech detection on the processed data. This segmentation improves speech detection accuracy by performing detection on cleaner, pre-processed signals rather than raw noisy input, while keeping each individual stage relatively simple.
Solution Approach 2:
The first stage of interference cancellation serves as a preliminary action that prepares the audio data for more accurate speech detection in the second stage. By removing noise and echo before detection, the system creates better conditions for accurate speech identification, resolving the contradiction between processing simplicity and detection accuracy.
Data Source
AI summary
Techniques for improving adaptive interference cancellation (AIC) using cascaded AIC algorithms are described. To improve an accuracy of detecting speech, a device may perform a first stage of AIC to generate isolated audio data and may generate speech mask data indicating time windows when speech is detected in the isolated audio data. Based on the speech mask data, the device may perform second AIC to generate output audio data, with adaptation of the adaptive filter enabled when the speech is not detected and disabled when the speech is detected. Thus, the first AIC improves the accuracy with which the device detects that speech is present and the second AIC reduces distortion in the output audio data by not updating filter coefficient values when the speech is present. The first AIC may use playback audio data, microphone audio data or beamformed audio data as reference signals.