Time-Domain Noise Reduction for Voice Signals

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing noise cancellation methods in high noise environments, such as automotive settings, require computationally intensive transforms like FFT or DCT for spectral subtraction, making them costly and inefficient for voice signal processing.

Innovation Solution

A time-domain noise reduction method that estimates noise levels and adjusts noise reduction parameters based on the presence of speech, eliminating the need for spectral transforms and minimizing speech distortion, suitable for use in acoustic echo cancellers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If spectral subtraction with FFT or DCT transforms is used for noise reduction, then noise reduction effectiveness is improved, but computational complexity and processing cost increase significantly

Engineering Contradiction:

Improvenoise reduction effectivenessVSAvoidcomputational complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces the mechanical transform-based spectral subtraction system with a time-domain statistical model. Instead of using FFT/DCT transforms to separate speech and noise in frequency domain, the invention uses autoregressive (AR) modeling and autocorrelation analysis directly in the time domain to estimate noise characteristics and subtract noise from speech signals, eliminating computationally intensive transforms while maintaining noise reduction effectiveness

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameters of noise reduction by transitioning from frequency-domain spectral parameters to time-domain statistical parameters. The method uses autocorrelation lags and AR model coefficients as parameters instead of frequency spectrum parameters, enabling noise reduction through time-domain statistical analysis rather than frequency-domain spectral manipulation

Inventive Principle:

Principle #35Parameter changes

2Reliability

If aggressive noise reduction is applied to remove background noise, then noise reduction effectiveness is improved, but speech distortion increases

Engineering Contradiction:

Improvenoise reduction effectivenessVSAvoidspeech distortion

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent implements dynamic noise reduction by continuously tracking speech presence using energy ratio thresholds and adapting the noise estimation process accordingly. The system dynamically switches between noise-only estimation and speech-plus-noise estimation modes based on real-time signal analysis, allowing aggressive noise reduction when only noise is present and gentle processing when speech is detected, thereby reducing speech distortion while maintaining noise reduction effectiveness

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent uses feedback mechanisms to monitor the output signal quality and adjust noise reduction parameters accordingly. By continuously analyzing the energy ratios and comparing against thresholds, the system provides feedback to control the aggressiveness of noise reduction, ensuring that speech components are preserved while background noise is suppressed

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS7908138B2Low complexity noise reduction method

Publication Date: 2011.03.15 IP GEM GRP LLC

AI summary

To reduce noise in an input signal that may contain speech, first an estimate of the noise level in the signal is obtained. The level of the input signal is then compared with the noise level estimate signal to determine whether speech is dominant. Less aggressive noise reduction is applied to the input signal when speech is dominant than when only noise is present.