Demonstration

Frame-based space-time covariance matrix estimation for PEVD-based speech enhancement

This animation shows the beampatterns obtained through polynomial eigenvalue (PEVD) decomposition of the frame-based space-time covariance matrix estimate obtained following the procedure in [1].

The desired speech source is located at 90° and is corrupted by diffuse speech-shaped noise at 0 dB SNR.

Several values of α are plotted, with larger values of α showing smaller beam fluctuations with speech activity.

GIF of beampatterns

References

[1]   E. d’Olne, V. W. Neo, and P. A. Naylor, Frame-based space-time covariance matrix estimation for polynomial eigenvalue decomposition-based speech enhancement, in Proc. IEEE Int. Workshop on Acoust. Signal Enhancement (IWAENC), Bamberg, Germany, 2022. s