Cross-Tower Echo Cancellation for Phase-Aware Voice Enhancement

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional echo cancellation methods using deep learning techniques in the frequency domain fail to accurately reflect the phase of input signals, leading to suboptimal performance in removing acoustic echoes.

Innovation Solution

An integrated noise and echo signal removal device using a parallel deep neural network structure that separately estimates noise and echo signals using multiple sequentially connected artificial neural networks, applying an attention mechanism to enhance the accuracy of signal estimation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional deep learning methods cancel acoustic echo in the frequency domain, then echo cancellation is performed using complex values, but the phase of the input signal is not directly reflected leading to deteriorated echo cancellation performance

Engineering Contradiction:

Improveecho cancellation performanceVSAvoidphase information loss

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The patent segments the echo cancellation process into separate estimation stages for different signal components (echo signal, noise signal, and speech signal) using multiple neural networks. This segmentation allows each network to focus on specific signal characteristics, preserving phase information for the echo signal while handling other components separately, thereby resolving the contradiction between echo cancellation performance and phase information retention.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If a single neural network is used for signal estimation, then the structure is simple, but the accuracy of separate noise and echo signal estimation is insufficient

Engineering Contradiction:

Improvesignal estimation accuracyVSAvoidneural network structure complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the signal estimation task into multiple specialized neural networks: a first neural network for echo signal estimation, a second neural network for noise signal estimation, and a third neural network for speech signal estimation. This segmentation improves estimation accuracy for each signal type while managing complexity through modular architecture where each network focuses on specific signal characteristics.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent merges the outputs of multiple specialized neural networks to produce the final enhanced speech signal. The estimated echo, noise, and speech signals from separate networks are combined through addition and subtraction operations, achieving high overall accuracy by leveraging the strengths of each individual network while maintaining a manageable system structure.

Inventive Principle:

Principle #5Merging (Combining)

3Measurement precision

If multiple sequentially connected artificial neural networks are used to separately estimate noise and echo signals, then the accuracy of signal estimation is improved, but the computational complexity increases

Engineering Contradiction:

Improvenoise and echo signal estimation accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by stationary object

Solution Approach 1:

The patent segments the computational workload across multiple specialized neural networks, where each network processes specific signal components in parallel rather than sequentially. This segmentation improves estimation accuracy for each signal type while optimizing computational energy consumption by avoiding redundant processing and allowing efficient parallel computation of separate signal estimates.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12526368B2Learning method for integrated noise echo cancellation system using cross-tower network

Publication Date: 2026.01.13 INDUSTRY UNIVERSITY COOPERATION FOUNDATION HANYANG UNIVERSITY

AI summary

An integrated noise and echo signal removal device using parallel deep neural network according to an embodiment comprises a microphone encoder that receives a microphone input signal including an echo signal, and a speaker's voice signal, converts the microphone input signal into first input information, and outputs the converted first input information, a far-end signal encoder that receives a far-end signal, converts the far-end signal into second input information, and outputs the converted second input information, a pre-learned second artificial neural network having a third input information, which is the sum of the first input information and the second input information, as input information, and having an estimated echo signal obtained by estimating the echo signal from the second input information as output information, a pre-learned third artificial neural network having the third input information as input information and having an estimated noise signal obtained by estimating the noise signal from the second input information as output information, a voice signal estimator configured to output an estimated voice signal obtained by estimating the voice information based on the estimated echo signal, the estimated noise echo signal, and the second input information.