Model-Based Noise PSD Estimation from Speech in Non-Stationary Noise

Abstract

Most speech enhancement algorithms need an estimate of the noise power spectral density (PSD) to work. In this paper, we introduce a model-based framework for doing noise PSD estimation. The proposed framework allows us to include prior spectral information about the speech and noise sources, can be configured to have zero tracking delay, and does not depend on estimated speech presence probabilities. This is in contrast to other noise PSD estimators which often have a too large tracking delay to give good results in nonstationary situations and offer no consistent way of including prior information about the speech or the noise type. The results show that the proposed method outperforms state-of-the-art noise PSD estimators in terms of tracking speed and estimation accuracy.

A Study on How Pre-whitening Influences Fundamental Frequency Estimation

“…Larger differences occuring when P is high, for the babble and restaurant noise types, implies that even if a better noise PSD spectrum could be captured (since a lower GER could be achieved), the conventional noise PSD estimators do not react quickly to nonstationary noise conditions and, therefore, the estimated noise PSD spectrum does not correctly fit the true one. This suggests a future improvement of prewhitening, for example based on codebook based approach [25,26], which can better encompass the noise characteristics. Based on this, we did not select a very high value of P for the previous experiment.…”

On Optimal Filtering for Speech Decomposition

“…It is important to mention that babble noise is one of the most difficult noise types to remove, since it is highly nonstationary and contains similar spectral content to speech. In this paper, we considered the noise statistics estimated with the default settings of the minimum statistics approach [17], but in a future improvement, the developed principle herein can be combined with a codebook-based approach [30], in order to get better estimates of the noise statistics. With respect to the Itakura-Saito distance (ISD), the ISD of the voiced component obtained by the optimal filtering formulation is lower than the other methods.…”

Model-Based Voice Activity Detection in Wireless Acoustic Sensor Networks

“…The traditional noise estimation methods, such as the minimum mean-square error (MMSE) noise PSD estimator [9], [10] and the minimum statistics (MS) noise PSD estimator [11] are widely used, but these methods have limited performance when dealing with non-stationary noise. In [12], the authors introduced a model-based noise PSD estimator by applying a statistical model to the speech and noise signals. The proposed noise PSD estimator is able to include prior spectral information about speech and different types of non-stationary noise [13].…”

“…In this section, we briefly summarize a model-based noise PSD estimation method which are able to track non-stationary noise, such as babble noise. A detailed description can be found in [12].…”

“…The definitions of f ,v 0 m,l and⌃ l can be found in [12]. A more detailed derivation of the model-based noise PSD estimation is available in [12]. The estimated speech PSD can be obtained in a similar way.…”