The paper presents a novel method for estimating the power spectral density (PSD) of nonstationary noise from noisy speech signals. Unlike traditional methods that use voice activity detectors (VADs), this approach tracks spectral minima in each frequency band without distinguishing between speech activity and pause. The method minimizes a conditional mean square estimation error criterion to derive an optimal smoothing parameter for recursive smoothing of the noisy speech signal's PSD. An unbiased noise estimator is developed based on the optimally smoothed PSD estimate and the analysis of spectral minima statistics. The estimator is suitable for real-time implementations and includes a bias compensation algorithm to improve performance in nonstationary noise. The paper also introduces a method to speed up the tracking of spectral minima. Experimental results show that the proposed method outperforms traditional VAD-based approaches in speech enhancement and low-bit-rate speech coding, particularly in preserving weak speech sounds and improving intelligibility.The paper presents a novel method for estimating the power spectral density (PSD) of nonstationary noise from noisy speech signals. Unlike traditional methods that use voice activity detectors (VADs), this approach tracks spectral minima in each frequency band without distinguishing between speech activity and pause. The method minimizes a conditional mean square estimation error criterion to derive an optimal smoothing parameter for recursive smoothing of the noisy speech signal's PSD. An unbiased noise estimator is developed based on the optimally smoothed PSD estimate and the analysis of spectral minima statistics. The estimator is suitable for real-time implementations and includes a bias compensation algorithm to improve performance in nonstationary noise. The paper also introduces a method to speed up the tracking of spectral minima. Experimental results show that the proposed method outperforms traditional VAD-based approaches in speech enhancement and low-bit-rate speech coding, particularly in preserving weak speech sounds and improving intelligibility.