File Exchange

image thumbnail

Noise tracking algorithm for single-microphone speech signals

version 1.5 (573 KB) by

Implements a data-driven noise tracking algorithm suitable for nonstationary noise sources.



View License

Contains m-files to train and implement the MMSE noise tracking algorithm described in:
J.S. Erkelens and R. Heusdens, "Tracking of nonstationary noise based on data-driven recursive noise power estimation", IEEE Trans. Audio, Speech & Lang. Proc., Vol. 16, No. 6, pp. 1112-1123, August 2008.

See Description.doc in the zip-file.

Comments and Ratings (5)

Kamil Wojcicki

Kamil Wojcicki (view profile)

Jan Erkelens

I guess you mean the train data? I used a lot of train data, which I cannot upload. You can either use your own train data, or use the pretrained gainfunctions I have included, in which case no training is required. See the example at the end of Description.doc. The m-files I used for training were added to assist people who want to do their own training.


Why submit files to the FEX that refer to data files on your hard drive? Perhaps your going to update this submission with your computer login details? ;-)


JEON (view profile)



Bug fix in createDFTAmatrix.m


The safety net in NoiseTracker.m assumed a sampling frequency of 8 kHz. NoiseTracker.m now reads the noisy speech from a .wav-file and the sampling frequency is automatically detected.


- Removed potsband.m (available online elsewhere, see Description.doc)
- Added pdf of the paper + IEEE copyright statement


Added an example to Description.doc and explained how the noise tracker may be used at different sampling frequencies.


The variable names in GainTables.mat have been changed to match those used in NoiseTracker.m, that is
G1 => G_A
G2 => G_A2
Gd2new => G_D2

MATLAB Release
MATLAB 7.5 (R2007b)

Download apps, toolboxes, and other File Exchange content using Add-On Explorer in MATLAB.

» Watch video