Signal Processing Device Linear Nonlinear Echo Cancellation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing echo cancellation technologies are insufficient in handling noise independent from reproduction sound volume, such as housing vibration or speaker distortion, leading to distortion in user utterances.

Innovation Solution

A signal processing device and method that includes a linear echo canceller followed by a non-linear echo suppressor using machine learning to further suppress residual echoes and noise, employing a deep neural network to generate a second suppression signal from a first suppression signal obtained through linear processing of mixed sound signals collected by a microphone.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If linear processing is used for echo cancellation, then echo components can be suppressed, but noise independent from reproduction sound volume (housing vibration, speaker distortion) cannot be effectively handled

Engineering Contradiction:

Improveecho cancellation performanceVSAvoidability to handle different noise types

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The echo cancellation system is segmented into two distinct processing stages: linear processing for handling linear echo components and non-linear processing for handling non-linear noise components. This segmentation allows each processing stage to be optimized for its specific type of noise, thereby improving overall echo cancellation performance and adaptability to different noise types.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system changes the processing parameter from linear to non-linear by applying different processing methods to different noise components. Linear processing parameters are used for echo cancellation while non-linear processing parameters are used for handling housing vibration and speaker distortion, enabling the system to adapt to various noise characteristics.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If non-linear processing is applied to suppress residual echoes and noise, then echo cancellation performance is improved, but calculation resources and model size may increase

Engineering Contradiction:

Improveecho cancellation performanceVSAvoidmodel size and calculation resources

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system applies non-linear processing only to the residual echo and noise components that remain after linear processing, rather than applying it to the entire signal. This partial action approach maintains improved echo cancellation performance while reducing the calculation resources and model size required compared to applying non-linear processing throughout the entire signal processing chain.

Inventive Principle:

Principle #16Partial or excessive action

3Reliability

If multiple nonlinear components are processed separately, then each component can be handled specifically, but device complexity and processing overhead increase

Engineering Contradiction:

Improvenoise suppression accuracyVSAvoidprocessing structure complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

Multiple non-linear noise components (housing vibration, speaker distortion, and residual echoes) are merged into a single non-linear processing stage. This combining approach allows the system to handle various non-linear noise types effectively while maintaining low device complexity and avoiding the overhead of separate processing structures for each noise component.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS12148442B2Signal processing device and signal processing method

Publication Date: 2024.11.19 SONY GROUP CORP

AI summary

There is provided a signal processing device that includes a processing unit configured to collect, through a microphone, a mixed sound signal as a mixed sound of a speaker reproduction signal and a target signal, input a first suppression signal resulting from suppression of the speaker reproduction signal from the mixed sound signal by linear processing, the speaker reproduction signal, and the mixed sound signal collected through the microphone, and output a second suppression signal resulting from further suppression of the speaker reproduction signal from the first suppression signal by non-linear processing.