SiSEC 2010: Source separation in the presence of real-world background noise (1 source with 2 or 4 channels)

Jani Even*, Hiroshi Saruwatari and Kiyohiro Shikanoi

Nara Institute of Science and Technology, JAPAN

(e-mails: even@is.naist.jp sawatari@is.naist.jp)

We proposed a method for the extraction of a dominant speech from a mixture (Noise cancellation).
The method can be divided in 4 main steps
(1) The blind signal extraction (BSE) method in [2] is used to obtain a single channel estimate of the target speech.
(2) This speech estimate is cancelled to get the estimate of the noise spatial image [3] (linear processing).
(3) A post filter (shaping of the spectrum) is used to suppress the estimated noise spatial image from the observed signal (channel wise non linear processing). The minimum value of the post filter shaping the spectrum is set by using the noise only period (determined using the ratio between noise estimate and observation).
(4) The processed channels are merged together with a delay and sum (DS) beamformer in the estimated direction of arrival (DOA) of the target speech
The DOA estimation is done using the estimated steering vector (higher frequencies where spatial aliasing occurs are ignored).

----------------------------------------------
References:
[1] Y. Takahashi, T. Takatani, K. Osako, H. Saruwatari, and
K. Shikano, ``Blind spatial subtraction array for speech
enhancement in noisy environment,'' IEEE Transactions on Audio,
Speech and Language Processing, vol.17, no.4, pp.650--664, 2009.

[2] J. Even, H. Saruwatari, K. Shikano, T. Takatani, ``SPEECH ENHANCEMENT IN PRESENCE OF DIFFUSE BACKGROUND NOISE: WHY USING BLIND SIGNAL EXTRACTION?,'' Proc. ICASSP 2010, pp. 4770--4773, Dallas, U.S.A., March 2010.

[3] J. Even, H. Saruwatari and K. Shikano, ``Blind Signal Extraction Based Speech Enhancement in Presence of Diffuse Background Noise,'' 2009 International Workshop on Statistical Signal Processing (SSP2009), pp.513--517, 2009.