Noise Power Spectral Density Estimation

Here we give Audio examples and the code for the proposed noise power spectral density estimator based on a speech presence probability estimator with fixed priors. The algorithm is proposed in the following papers:

• Timo Gerkmann and Richard C. Hendriks, "Unbiased MMSE-based Noise Power Estimation with Low Complexity and Low Tracking Delay", IEEE Trans. Audio, Speech and Language Processing, 2012.

• Timo Gerkmann and Richard C. Hendriks, "Noise Power Estimation Based on the Probability of Speech Presence", IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, New Paltz, NY, USA, Oct. 2011.

Audio examples

These audio examples compare single channel noise reduction using Martin's Minimum Statistics [1], the bias compensated MMSE approach [2], and the proposed SPP approach.
The proposed approach is computationally more efficient than the Minimum Statistics [1] or MMSE-BC [2] (approximately factor 4.5 in Matlab) and also requires less memory.

Wiener filter
Noise type	noisy	proposed	Minimum Statistics [1]	MMSE-BC [2]
modulated white	audio	audio	audio	audio
white	audio	audio	audio	audio
traffic	audio	audio	audio	audio
babble	audio	audio	audio	audio

Super-Gaussian filter from [3] with γ=1, ν=0.6
Noise type	noisy	proposed	Minimum Statistics [1]	MMSE-BC [2]
modulated white	audio	audio	audio	audio
white	audio	audio	audio	audio
traffic	audio	audio	audio	audio
babble	audio	audio	audio	audio

[1] R. Martin, "Noise power spectral density estimation based on optimal smoothing and minimum statistics," IEEE Transactions on Speech and Audio Processing, vol. 9, no. 5, pp. 504-512, Jul. 2001.

[2] R. C. Hendriks, R. Heusdens, and J. Jensen, "MMSE based noise PSD tracking with low complexity," IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 4266-4269, Mar. 2010.

Code

The code for the proposed approach can be found here: [Download]

The code for the MMSE-BC approach [2] can be found here.