Evaluation of noise estimation algorithms based on minimum statistics and signal to noise ratio

Abstract

© 2016 IEEE. The paper reports on the objective evaluation and comparison of the two noise estimation algorithms for noisy speech signals. Both algorithms are based on observation that local minima in noisy speech spectrogram are close to the power level of the noise signal. The first algorithm directly searches spectrogram for the local minima and those values use to update noise power spectrum density (psd). The second one updates noise psd continually, but the segments with sufficiently low signal to noise ration have higher influence. Mean and median square error as well as interquartile range of square error were used as objective measure. All evaluations were done on NOIZEUS database.