Academia.eduAcademia.edu

Noise estimation

description759 papers
group18 followers
lightbulbAbout this topic
Noise estimation is the process of quantifying the level of noise present in a signal or dataset, often used in signal processing and statistical analysis. It involves assessing the variance or power of noise to improve the accuracy of data interpretation, filtering, and reconstruction techniques.
lightbulbAbout this topic
Noise estimation is the process of quantifying the level of noise present in a signal or dataset, often used in signal processing and statistical analysis. It involves assessing the variance or power of noise to improve the accuracy of data interpretation, filtering, and reconstruction techniques.

Key research themes

1. How are statistical models used to characterize and estimate noise in digital images?

This theme focuses on the development and utilization of statistical noise models to characterize various sources of noise in digital images and to estimate noise parameters. It is fundamental for designing effective image denoising and quality assessment methods, as understanding noise statistical properties directly informs image processing algorithms. This area is critical because noise appears in multiple forms (Gaussian, Poisson, speckle, impulse) and arises from various stages such as image acquisition, sensor imperfections, and environmental effects.

Key finding: This paper provides a comprehensive quantitative analysis of various fundamental noise models in digital images, such as Gaussian noise (modeled by normal distributions with zero mean and known variance), white noise (modeled... Read more
Key finding: This study introduces a Bayesian probabilistic hierarchical model based on the theoretical imaging noise components, capable of decomposing image noise sources in sensors. The model accounts for variability across pixels and... Read more
Key finding: This paper innovatively applies the Rayleigh distribution combined with Monte Carlo estimation to model and detect image noise areas, particularly impulse noise, by statistically estimating the distribution of pixel... Read more
Key finding: The authors develop a statistical method to correct measured source levels by accounting for background noise contributions, emphasizing that noise corrections should increase with the steadiness of background noise rather... Read more
Key finding: This chapter discusses fundamental statistical modeling of noise in signals/ images, emphasizing stationary independent zero-mean white Gaussian noise for additive noise modeling. It explains how empirical noise realizations... Read more

2. How can uncertainty quantification improve noise estimation and speech enhancement models?

This research theme investigates the integration of uncertainty modeling—both aleatoric (data inherent) and epistemic (model parameter)—in neural network frameworks for noise estimation and speech enhancement. Accurately capturing uncertainty in noise estimates is shown to improve model reliability, robustness, and performance, especially under varying or previously unseen noise conditions. Quantification of uncertainty informs confidence in predictions and enables improved statistical inference, loss functions, and noise parameter estimation.

Key finding: The study proposes a neural network that simultaneously estimates the Wiener filter for speech enhancement and its associated uncertainty by modeling the posterior distribution of spectral coefficients via maximum a... Read more
Key finding: This paper develops a framework that jointly models aleatoric and epistemic uncertainties in deep neural network-based speech enhancement by estimating statistical moments of the speech posterior and employing Bayesian... Read more
Key finding: This paper presents a novel noise power spectral density (PSD) estimation algorithm using a derivative-based high-pass filter designed to track non-stationary noise dynamics in speech enhancement tasks. Unlike minimum... Read more

3. What are the methods and challenges in noise parameter estimation for environmental noise, especially traffic noise, using data-driven and modeling approaches?

This theme covers the environmental noise estimation domain, focusing primarily on traffic noise, where noise parameters are estimated using experimental monitoring systems and computational models. The challenges include capturing variabilities due to traffic volume, vehicle types, pavement surface, and environmental conditions. Data-driven techniques, such as particle filters and regression models, are employed to calibrate noise prediction parameters for accurate environmental impact assessments and urban noise mapping. Such methods enable informed urban planning and noise pollution mitigation strategies.

Key finding: This paper proposes a procedure to determine traffic noise model parameters adaptively using a particle filter algorithm that integrates vehicle velocity and type data from weigh-in-motion systems and ambient noise... Read more
Key finding: The authors develop a regressive predictive model of road traffic noise calibrated exclusively on computed noise levels independent of field measurements, utilizing multilinear regression over variables such as traffic flow,... Read more
Key finding: This review synthesizes noise monitoring and modeling studies worldwide, notably emphasizing the disproportionate focus on traffic noise (>90% of studies), and documents methods for 2D and 3D noise mapping supporting urban... Read more
Key finding: This study quantitatively assesses the uncertainty of equivalent sound pressure level measurements from a year-long stationary traffic noise monitoring station in urban Poland. Using high resolution sound and velocity... Read more

All papers in Noise estimation

This paper presents two methods for determining the noise power (mainly caused by crosstalk) in Digital Subscriber Line (DSL) networks. A fuzzy system approach is compared to a linear regression approach. Both are applied to a real world... more
In this paper, we present a simultaneous detection and estimation approach for speech enhancement in nonstationary noise environments. A detector for speech presence in the short-time Fourier transform domain is combined with an... more
We describe the on-board electronics chain and the on-ground data processing pipeline that will operate on data from the Herschel-SPIRE photometer to produce calibrated astronomical products. Data from the three photometer arrays will be... more
This paper study the blind estimation of th巴 diffuse back ground noise for the hands-free speech interface. Some recent papers showed that it is possible to use blind signal separation (BSS) to estimate the di仔use background noise by... more
We report on the sensitivity of SPIRE photometers on the Herschel Space Observatory. Specifically, we measure the confusion noise from observations taken during the science demonstration phase of the Herschel Multi-tiered Extragalactic... more
This work presents a statistical analysis of a class of jointly optimized beamformer-assisted acoustic echo cancelers (AEC) with the beamformer (BF) implemented in the Generalized Sidelobe Canceler (GSC) form and using the least-mean... more
This work presents a statistical analysis of a class of jointly optimized beamformer-assisted acoustic echo cancelers (AEC) with the beamformer (BF) implemented in the Generalized Sidelobe Canceler (GSC) form and using the least-mean... more
Given a Brownian Motion $W$, in this paper we study the asymptotic behavior, as $\eps \to 0$, of the quadratic covariation between $f (\eps W)$ and $W$ in the case in which $f$ is not smooth. Among the main features discovered is that the... more
An adaptation algorithm for the parameters selection of the multiple frequency hybrid observer in [1] is proposed. The algorithm is tailored for numerical implementation and is aimed to retrieve, exploiting a suitable amount of past data,... more