INTERSPEECH 2007
8th Annual Conference of the International Speech Communication Association

Antwerp, Belgium
August 27-31, 2007

DFT Domain Subspace Based Noise Tracking for Speech Enhancement

Richard C. Hendriks, Jesper Jensen, Richard Heusdens

Delft University of Technology, The Netherlands

Most DFT domain based speech enhancement methods are dependent on an estimate of the noise power spectral density (PSD). For non-stationary noise sources it is desirable to estimate the noise PSD also in spectral regions where speech is present. In this paper a new method for noise tracking is presented, based on eigenvalue decompositions of correlation matrices that are constructed from time series of noisy DFT coefficients. The presented method can estimate the noise PSD at time-frequency points where both speech and noise are present. In comparison to state-of-the-art noise tracking algorithms the proposed algorithm reduces the estimation error between the estimated and the true noise PSD and improves segmental SNR when combined with an enhancement system with several dB.

Full Paper

Acoustic Examples

noisy_train_5dB.wav   Speech signal degraded by noise originating from a passing train at a global SNR of 5 dB
Min_Stat_train_5dB.wav   Enhanced version of the signal "noisy_train_5dB.wav", based on minimum statistics (ref [3]). As enhancement gain function method proposed in ref [2] with \gamma=2 and \nu=0.1 is used.
DFT_Sub_Space_train_5dB.wav   Enhanced version of the signal "noisy_train_5dB.wav", based on proposed noise tracking algorithm. As enhancement gain function method proposed in ref [2] with \gamma=2 and \nu=0.1 is used.
noisy_nonstat_white_10dB.wav   Speech signal degraded by non-stationary white noise. The initial noise level is at an SNR of 10 dB. Then the noise level gradually increases in one second by 15 dB where it stays at that level for 2 seconds after which it decreases again by 15 dB in one second.
Min_Stat_nonstat_white_10dB.wav   Enhanced version of the signal "noisy_nonstat_white_10dB.wav", based on minimum statistics (ref [3]). As enhancement gain function method proposed in ref [2] with \gamma=2 and \nu=0.1 is used.
DFT_Sub_Space_nonstat_white_10dB.wav   Enhanced version of the signal "noisy_nonstat_white_10dB.wav", based on proposed noise tracking algorithm. As enhancement gain function method proposed in ref [2] with \gamma=2 and \nu=0.1 is used.

Bibliographic reference.  Hendriks, Richard C. / Jensen, Jesper / Heusdens, Richard (2007): "DFT domain subspace based noise tracking for speech enhancement", In INTERSPEECH-2007, 830-833.