Wideband Speech Enhancement With a Multi-Window Method for Spectrum Estimation

Conference: Sprachkommunikation 2006 - ITG-Fachtagung
04/26/2006 - 04/28/2006 at Kiel, Germany

Proceedings: Sprachkommunikation 2006

Pages: 4Language: englishTyp: PDF

Personal VDE Members are entitled to a 10% discount on this title

Janardhanan, Deepa; Heute, Ulrich (Institute for Circuit and System Theory, University of Kiel, Kaiserstr. 2, 24143 Kiel, Germany)

In this paper the enhancement of wideband speech is addressed, wherein emphasis is laid on the improvement of the noise-estimation and spectral-subtraction methods by employing a more accurate spectrum-estimation than the conventional periodogram. In a single-channel speech communication scenario, the enhancement of speech by means of spectral-subtraction is performed generally in the frequency domain. Therefore, the quality of the enhanced speech depends on the accuracy of the spectral estimates. An accurate spectrum estimate becomes even more important in the case of wideband speech enhancement. This is due to the inclusion of the high-variance speech components above 3400 Hz. A multi-window spectrum-estimation (MWSE) method is used instead of the conventional periodogram. It is observed that by employing the MWSE method for the power spectrum estimation the performance of the noise estimation and the spectral-subtraction is improved, thereby also enhancing the quality of the resulting speech.