Acoustic Echo Cancellation via Low-Frequency Double Talk Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Acoustic echo cancellation systems face challenges in accurately detecting double talk, leading to excessive residual echo or missed voice content due to inaccurate Double Talk Detection (DTD), especially in devices with small speakers that produce distorted low frequencies.

Innovation Solution

The system employs a high-pass filter to remove low-frequency content from the far-end audio and uses a least mean square (LMS) adaptive filter to detect near-end voice in the low-frequency sub-band, freezing or enabling filter adaptation based on the presence of low-frequency audio content, and sends comfort noise or original audio accordingly.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If Double Talk Detection is used to detect near-end voice during speaker activity, then near-end voice transmission is enabled, but inaccurate detection causes excessive residual echo or missed voice content

Engineering Contradiction:

Improveaccuracy of near-end voice detectionVSAvoidresidual echo and missed voice content

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent segments the audio frequency spectrum into different bands, specifically utilizing the low-frequency band (below the high-pass filter cutoff) to detect near-end voice. This segmentation allows the system to monitor a specific frequency range where near-end voice energy concentrates while far-end echo is suppressed, improving detection accuracy and reducing false positives that cause residual echo or missed voice content.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by focusing detection efforts on the low-frequency sub-band rather than analyzing the entire frequency spectrum. By concentrating computational resources and detection algorithms on this specific local frequency region where near-end voice is most prominent and far-end echo is minimized, the system achieves more reliable voice activity detection with reduced errors.

Inventive Principle:

Principle #3Local quality

2Volume of moving object

If small speakers are used in the device, then device size is reduced, but low-frequency production capability is compromised causing distortion

Engineering Contradiction:

Improvedevice sizeVSAvoidlow-frequency audio quality

Core Design Contradiction:

Volume of moving objectVSManufacturing precision

Solution Approach 1:

The patent extracts and removes the problematic low-frequency content from the far-end audio signal using a high-pass filter before it reaches the small speaker. By taking out the low-frequency components that small speakers cannot reproduce accurately, the system prevents distortion and intelligibility loss while maintaining the compact device form factor with small speakers.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent converts the limitation of small speakers (inability to produce low frequencies) into a benefit by using the same frequency characteristic for echo cancellation. The natural attenuation of low frequencies by small speakers reduces far-end echo energy in this band, which the system then exploits to improve near-end voice detection accuracy while maintaining compact dimensions.

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach effectively reduces echo cancellation errors by accurately distinguishing between near-end voice and far-end echo, ensuring clear transmission of near-end voice while minimizing residual echo.

Implementation Method 1

A high-pass filter is placed on the receive side to remove any low-frequency content

Methodology Applied

Scientific EffectHigh-pass filtering: Filter (electronic)

Implementation Method 2

The AEC system uses an adaptive filter to estimate the channel from the speaker to the microphone of the device

Methodology Applied

Scientific EffectAcoustic echo cancellation: Echo

Implementation Method 3

subtracting any high-pass filtered audio content from the audio content captured by the microphone using a least mean square (LMS) adaptive filter

Methodology Applied

Scientific EffectLeast mean square adaptive filtering:

Implementation Method 4

examining audio content captured by the microphone to detect the presence of audio content in a low-frequency sub-band

Methodology Applied

Scientific EffectLow-frequency voice detection:

Data Source

PatentUS10636435B1Acoustic echo cancellation using low-frequency double talk detection

Publication Date: 2020.04.28 MICROSEMI SEMICONDUCTOR US INC

AI summary

A method for canceling acoustic far-end audio echo content includes high-pass filtering audio content received from a far end and playing the high-pass filtered audio content through a speaker, examining audio content captured by a microphone to detect the presence of audio content in a low-frequency sub band after subtracting high-pass filtered audio content from the audio content captured by the microphone using a least mean square (LMS) adaptive filter. If audio content in a low-frequency sub band is detected in the audio content captured by the microphone, freezing adaptation of the LMS filter and sending to the far end the audio content captured by the microphone after subtracting, and if audio content in a low-frequency sub band is not detected in the audio content captured by the microphone, enabling adaptation of the LMS filter and sending to the far end the audio content captured by the microphone after subtracting.