Minimal distortion principle for blind source separation

Abstract

Abstnd: In blind source separation the numh:r of the m o r s is usually assumed to be equal to that of the sources. In this c a x a n i n w ' appears with which any lmcar transform of an estimated source signal can also be considered another estimation of the source signal. M m w in the case that the number of the msors is greater than that of the sou~ces, another indctmninacy arises due to the redundancy of the sensors. Although these indctaminacies arc often mnsidercd unsubstantial and have k n eliminated without defmite bass, an appropriate "alization of the separator is to enhance the acauacy of the separation resdlt parlicularly in the case of convolutive mixture. This papa shows two principles for eliminating these inddmninacie:i: (i) minimal distortion principle and (ii) inverse minimal distortion principle.

On separation performance enhancement in convolutive blind source separation

“…This approach accepts the filtering done by the mixing system without adding new distortions. In [14] a similar technique, known as the minimal distortion principle, has been proposed. This method uses the following unmixing matrix…”

Speech Enhancement via Combination of Wiener Filter and Blind Source Separation

“…However, beamforming algorithms require a-priory knowledge about the acoustic environment and the sources involved, or a large number of sensors are required for good performance. Another algorithm for source separation is blind source separation (BSS) [8,9]. In BSS source signals are estimated only based on the information of signals observed at each input channel.…”

Two-channel speech denoising through minimum tracking

“…Instead of this arbitrary filtering, it is desirable to obtain an estimate a 11 (v)S(v) which corresponds to the clean speech signal as observed at the microphone. One approach is to apply the minimal distortion principle [5] using the new unmixing matrix…”

“…â (v) was obtained in the frequency domain using (5). A frame length of 1024 samples was used and the cross spectral density matrix was computed by averaging over five neighbouring frames.…”

“…An unmixing matrix W ref (v) was first obtained following [1]. For a fair comparison, the minimal distortion principle (MDP) described in [5] We also measured the log spectral distortion between the clean speech signal and the enhanced signals, and the results are shown in Table 2. It can be seen that the proposed method results in lower distortion.…”