Dual-Stage Noise Reduction Architecture for Audio Signal Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing noise cancellation techniques, such as Spectral Subtraction and Voice Activity Detection, are ineffective in environments with high levels of noise, leading to degraded audio quality and increased error rates in speech recognition systems due to non-linear distortion and sensitivity drift between acoustic channels.

Innovation Solution

A multi-channel noise cancellation system that combines adaptive noise cancellation and single-channel noise reduction using omni-directional and directional microphones, with auto-balancing and beamforming to maintain signal fidelity and reduce non-linear distortion, employing adaptive finite impulse response filters and linear filtering to isolate desired audio from undesired audio.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If Spectral Subtraction is used to reduce noise in speech recognition algorithms, then noise reduction is achieved in low noise environments, but the error rate increases significantly when the magnitude of undesired audio becomes large

Engineering Contradiction:

Improvenoise reductionVSAvoiderror rate

Core Design Contradiction:

Object-affected harmful factorsVSReliability

Solution Approach 1:

The patent divides the noise cancellation task into two distinct stages: a first stage using adaptive noise cancellation with multiple acoustic channels to handle non-stationary noise, and a second stage using single-channel noise reduction to handle stationary noise. This segmentation allows each stage to be optimized for specific noise conditions, preventing the error rate increase that occurs when a single method is used across all noise levels.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system dynamically switches between different noise cancellation approaches based on the characteristics of the noise environment. The dual-stage architecture enables adaptive selection of processing methods, where the first stage handles high-magnitude non-stationary noise and the second stage refines the output for stationary noise, maintaining reliability across varying noise conditions.

Inventive Principle:

Principle #15Dynamics

2Object-affected harmful factors

If non-linear treatment of acoustic signal is applied to reduce noise, then noise reduction is achieved, but non-linear distortion of desired audio is introduced which disrupts feature extraction

Engineering Contradiction:

Improvenoise reductionVSAvoidnon-linear distortion

Core Design Contradiction:

Object-affected harmful factorsVSLoss of information

Solution Approach 1:

The patent replaces non-linear signal processing methods with linear filtering approaches in the second stage of noise cancellation. By using linear filters such as Wiener filters or minimum mean-square error filters, the system achieves noise reduction while preserving the linear relationship between input and output signals, thereby avoiding non-linear distortion that disrupts speech feature extraction.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Object-affected harmful factors

If multiple acoustic channels are used to receive acoustic signals, then noise cancellation capability is improved, but sensitivity drift between channels occurs which reduces accuracy

Engineering Contradiction:

Improvenoise cancellation capabilityVSAvoidchannel balance

Core Design Contradiction:

Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The patent incorporates an auto-balancing mechanism that continuously monitors and adjusts the sensitivity of multiple acoustic channels. This feedback system detects drift in channel sensitivity and applies corrective gain adjustments to maintain balanced channel responses, ensuring accurate noise cancellation performance over time and across varying environmental conditions.

Inventive Principle:

Principle #23Feedback

4Difficulty of detecting and measuring

If Voice Activity Detection is used to detect desired speech, then speech detection is achieved in quiet environments, but the system performs poorly when magnitude of undesired audio is large

Engineering Contradiction:

Improvespeech detectionVSAvoiddetection accuracy

Core Design Contradiction:

Difficulty of detecting and measuringVSReliability

Solution Approach 1:

The patent applies noise cancellation processing before voice activity detection and speech recognition. By pre-processing the acoustic signals to reduce noise content in both stages, the system improves the quality of input signals for detection algorithms, enabling reliable speech detection even when the magnitude of undesired audio is large.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS9633670B2Dual stage noise reduction architecture for desired signal extraction

Publication Date: 2017.04.25 SOLOS TECH LTD

AI summary

Systems and methods are described to reduce undesired audio. An adaptive noise cancellation unit receives a main signal and a reference signal. The main signal has a main signal-to-noise ratio; the reference signal has a reference signal-to-noise ratio. The reference signal-to-noise ratio is less than the main signal-to-noise-ratio. The adaptive noise cancellation unit reduces undesired audio from the main signal. An output signal from the adaptive noise cancellation unit is input to a single channel noise cancellation unit. The single channel noise cancellation unit further reduces undesired audio from the output signal to provide mostly desired audio. A filter control creates a control signal from the main signal and the reference signal to control filtering in the adaptive noise cancellation unit and to control filtering in the single channel noise cancellation unit.