Sub-band Noise Estimator Using Posteriori Probability Maximization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise estimating methods in noise suppressors face issues with instability and rapid variation in noise power estimation, leading to unpleasant auditory sensations and distortion of enhanced speech, particularly when dealing with rapidly changing noise levels and non-stationary noise.
Innovation Solution
A noise estimation apparatus and method that utilize a sub-band noise estimator with a power calculator, a probability model holder, and an a posteriori probability maximizer to calculate an instantaneous noise power estimate based on the input power, estimated noise power, and a probability model that maximizes the posteriori probability of the noise power, incorporating stationarity models and a posteriori signal-to-noise ratio information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional noise estimating methods average input spectra within speech absent periods, then the noise estimation is simple to implement, but the noise power estimation becomes unstable and distorts enhanced speech when speech active periods are misidentified
Solution Approach 1:
The patent replaces the conventional mechanical averaging method with a statistical maximum likelihood estimation approach. Instead of simply averaging spectra during identified noise periods, the system uses probabilistic models to estimate noise power continuously, substituting the deterministic mechanical process with a statistical inference process that handles uncertainty in speech activity detection.
Solution Approach 2:
The patent changes the estimation parameter from discrete noise-period averaging to continuous maximum likelihood estimation. By transforming the estimation approach from time-domain averaging to frequency-domain statistical inference using probability density functions, the system achieves more stable noise power estimates that are less sensitive to speech activity detection errors.
2Productivity
If noise is estimated only in noise periods, then the estimation process is computationally efficient, but the noise power does not track noise variation during long speech active periods
Solution Approach 1:
The patent performs preliminary actions by continuously estimating noise power during both speech active and inactive periods rather than waiting for noise periods. The maximum likelihood estimation framework allows the system to proactively track noise variations at all times, preparing accurate noise power estimates before they are needed for enhancement, thus eliminating the tracking delay inherent in conventional methods.
3Measurement precision
If conventional methods use speech absent periods for noise estimation, then the noise estimation avoids speech contamination, but misidentification of speech active periods causes speech to be included in noise estimation
Solution Approach 1:
The patent introduces a probabilistic model as an intermediary between the observed input spectra and the noise power estimation. Instead of directly using speech absent period identification, the system employs maximum likelihood estimation with probability density functions that inherently handle the uncertainty of speech presence, acting as a mediator that reduces the impact of VAD errors on noise estimation purity.
4Measurement precision
If noise estimation is performed continuously during speech active periods, then the noise power tracks noise variation, but the estimation becomes sensitive to speech components and produces rapid variations
Solution Approach 1:
The patent applies partial action by using the maximum likelihood estimation framework that selectively incorporates information from the input spectra based on the probabilistic model. The system performs continuous estimation but only to the extent necessary to track noise variations, using the probability density function to filter out excessive speech-related variations while maintaining responsiveness to actual noise changes.
Data Source
AI summary
A noise estimation apparatus of estimating a noise in an input signal includes a sub-band noise estimator estimating a noise in a sub-band input signal, obtained by dividing the input signal by sub-bands. The sub-band noise estimator includes a power calculator calculating a sub-band input power of the sub-band input signal; a probability model holder holding information on probability model; and an a posteriori probability maximizer calculating an instantaneous estimated value of a sub-band noise power based on the sub-band input power, an estimated value of the sub-band noise power and the information on the probability model, so as to maximize a posteriori probability of the sub-band noise power. The information on the probability model includes a likelihood function regarding a posteriori signal-to-noise ratio (SNR) in dependence upon predictive a posteriori SNR; and a priori probability of the a posteriori SNR under a condition establishing averaged a posteriori SNR.