Speech Signal Processing Using Vibration Cues for Noise-Robust Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing speech signal processing systems struggle with robustness in separating user speech from mixed signals that include ambient noise and other non-user speech, limiting the effectiveness of speech recognition and voiceprint authentication.

Innovation Solution

A method that utilizes user-generated vibration and brain wave signals in conjunction with speech signals to enhance voiceprint recognition, leveraging these additional signals to improve confidence levels and suppress noise interference.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If voice signal processing is performed without voice activity detection, then processing continuity is maintained, but energy consumption increases and unnecessary processing occurs

Engineering Contradiction:
Improveenergy consumptionVSAvoidprocessing efficiency
Core Design Contradiction:
Loss of energyVSProductivity

Solution Approach 1:

The voice activity detection module performs preliminary detection on the voice signal before the main processing pipeline. By detecting voice activity in advance, the system determines whether subsequent processing operations are necessary, thereby avoiding unnecessary processing and reducing energy consumption while maintaining processing efficiency when voice is present

Inventive Principle:
Principle #10Preliminary action

2Reliability

If traditional voice activity detection methods are used, then voice activity can be detected, but processing delays occur and real-time performance deteriorates

Engineering Contradiction:
Improvevoice activity detection accuracyVSAvoidprocessing delay
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The voice activity detection is segmented into multiple parallel frequency bands instead of processing the entire spectrum sequentially. The signal is divided into different frequency ranges, and detection is performed simultaneously on each band, which reduces overall processing delay while maintaining detection accuracy through multi-band analysis

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs partial processing by focusing detection efforts only on frequency bands where voice activity is likely to occur, rather than uniformly processing all frequency bands. This selective approach reduces processing time while maintaining reliable detection of actual voice activity

Inventive Principle:
Principle #16Partial or excessive action

3Productivity

If simple voice activity detection is implemented, then processing speed is maintained, but detection accuracy and reliability decrease

Engineering Contradiction:
Improveprocessing speedVSAvoidvoice activity detection precision
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The detection system transitions from single-dimensional time-domain analysis to multi-dimensional analysis by incorporating frequency domain information through short-time Fourier transform. This adds a frequency dimension to the detection process, enabling more precise voice activity detection while maintaining processing speed through efficient spectral analysis

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

4Reliability

If multiple processing modules are cascaded, then processing comprehensiveness is improved, but system complexity and latency increase

Engineering Contradiction:
Improveprocessing completenessVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The voice activity detection module acts as a preliminary gatekeeper before the main processing pipeline. By performing detection in advance and controlling the flow of subsequent processing based on detection results, the system achieves comprehensive processing when needed while avoiding unnecessary processing stages, thereby reducing overall system complexity and latency

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4141867B1Voice signal processing method and related device therefor
Publication Date: 2026.05.20 HUAWEI TECH CO LTD
  • EP4141867B1 patent drawingFigure 1a
  • EP4141867B1 patent drawingFigure 1b(a)
  • EP4141867B1 patent drawingFigure 1b(b)

AI summary

A speech signal processing method and a related device thereof are provided. The method may be applied to the audio field and includes: obtaining a user speech signal captured by a sensor; obtaining a corresponding vibration signal when a user generates a speech, where the vibration signal indicates a vibration feature of a body part of the user, and the body part is a part that vibrates correspondingly based on sound-making behavior when the user is making a sound; and obtaining target speech information based on the vibration signal and the user speech signal captured by the sensor. In this application, the vibration signal is used as a basis for speech recognition. Because the vibration signal does not include an external non-user speech mixed during complex acoustic transmission, and is slightly affected by other ambient noise (for example, affected by reverberation), this part of noise interference can be well suppressed, and good speech recognition effect can be achieved.