Digital Signal Processing Reference
In-Depth Information
small value (approximate zero). In our simulation study, the convergent
threshold is set to be 0.0098. Correspondingly, the order in the
reconstruction of voiced speech signal is chosen to be 9 with around 7.25dB
maximum SNR, while the order of unvoiced speech signal is chosen to be 13
with around 5.01dB which is close to the maximum SNR. It is clear that the
maximum of the SNR curve occurs at the selected order determined by the
FCF although there is a slight bias against unvoiced speech signal due to its
noise like characteristics.
Figure 11-1.
Reconstruction of the Voiced Speech Signal.
5.2
Evaluation of Enhancement Performance
The performance evaluation of the proposed algorithm has been tested
and compared with the truncated SVD (TSVD) algorithm for MV estimation
reported in [7]. The fixed order of the TSVD algorithm is selected to be 14
by which the best enhanced performance is attained as described in [7]. Four
