Noise Estimation Apparatus Using Likelihood Maximization for Non-Stationary Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise estimation methods do not calculate non-speech prior probability, non-speech posterior probability, and noise signal variance based on the likelihood maximization criterion, leading to suboptimal noise estimation and poor performance in canceling non-stationary noise.
Innovation Solution
A noise estimation apparatus that uses weighted addition of log likelihoods from Gaussian distribution models in speech and non-speech segments, combined with posterior probabilities, to estimate noise signal variance based on the likelihood maximization criterion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional noise estimation methods are used, then the processing is simple, but the noise estimation precision is suboptimal and cannot effectively track non-stationary noise
Solution Approach 1:
The patent changes the fundamental parameter of noise estimation from rule-based variance calculation to likelihood maximization-based variance calculation. By using the likelihood maximization criterion to estimate noise signal variance σv,i2 instead of conventional recursive averaging, the system achieves optimal noise estimation precision while adapting to non-stationary noise characteristics.
Solution Approach 2:
The patent implements feedback by using estimated noise variance to update the likelihood function in subsequent iterations. The noise estimation unit continuously refines the noise signal variance estimate by feeding back the previously estimated variance σv,i-12 and comparing it with current observed signal statistics, enabling adaptive tracking of non-stationary noise.
2Reliability
If rule of thumb parameter adjustment is used, then the method is easy to implement, but the noise cancellation performance is poor
Solution Approach 1:
The patent performs preliminary action by pre-defining the likelihood function form and estimation framework before actual noise estimation. The system prepares the probabilistic model structure with speech and non-speech segment likelihoods, then executes optimal parameter estimation through likelihood maximization, ensuring reliable noise cancellation from the outset.
Solution Approach 2:
The patent substitutes the mechanical rule-based variance adjustment system with a probabilistic likelihood maximization system. Instead of mechanically applying fixed rules for variance update, the system uses statistical likelihood functions to automatically determine optimal noise variance estimates, replacing deterministic mechanical processing with probabilistic optimization.
Data Source
AI summary
A noise estimation apparatus which estimates a non-stationary noise component on the basis of the likelihood maximization criterion is provided. The noise estimation apparatus obtains the variance of a noise signal that causes a large value to be obtained by weighted addition of the sums each of which is obtained by adding the product of the log likelihood of a model of an observed signal expressed by a Gaussian distribution in a speech segment and a speech posterior probability in each frame, and the product of the log likelihood of a model of an observed signal expressed by a Gaussian distribution in a non-speech segment and a non-speech posterior probability in each frame, by using complex spectra of a plurality of observed signals up to the current frame.