Spectral Noise Estimation: A Python 3 Implementation of the Minimum Statistics Estimation

ABSTRACT

Noise estimation has been used majorly in imaging processing and voice speech recognition applications. Therefore, researchers have found optimal solutions to non-stationary noise estimation. Particularly, there is a proposed method that estimates spectral noise in a noisy speech signal which is based on two observations; speech pauses and approximation of power spectral densities of the noisy signal to the true noise during speech pauses. Though from recent studies, the observations obtained cannot be inferred for other types of signals especially RF signals and have not been tested on signals in the frequency domain, this paper bridges that gap of research and presents the results, analysis, and conclusion on the findings concerning the noise estimation with RF signals using an extension of the proposed method in the frequency domain. It presents a detailed methodology of implementation of the minimum statistics method for noise estimation in python 3 code which was tested with RF signals and thus met the requirement of dynamic thresholding with spectrum occupancy measurement.

Keywords: Spectral estimation, Minimum statistics, Smoothing, Noise estimation, RF signals, Dynamic thresholding

Download full-text

This work is licensed under a Creative Commons Attribution 4.0 International License.