Digital Signal Processing Reference
In-Depth Information
1.0
Error Function
Threshold Function
0.8
0.6
0.4
0.2
0.0
0.0
2.0
4.0
6.0
8.0
10.0
Decision Bands
Figure 6.33
The error and threshold functions for one frame
Split-Band Mixed Voicing
Correct estimation of the threshold level for each band is the most important
stage in MBE mixed-voicing estimation. The other important factor is that
more than one bit will be needed during the coding of the mixed-voicing
decision estimate. Since each band will require one bit, more bands will mean
higher accuracy but an increased bit rate. When closely examined however,
we see that if a spectrum contains an unvoiced band between two voiced
bands, the unvoiced signal in the middle is usually relatively small and if
it is declared as voiced, subjectively it would not make much difference.
This is very important because it saves bits when coding the mixed voicing-
decision. In this case a single point in the frequency spectrum can be used to
identify the voiced (low frequency) and unvoiced (high frequency) regions.
There may be several ways to obtain the single frequency marker or cut-off
point which separates the voiced and unvoiced parts. For example, using
MBE mixed-voicing, above, one can assume that the spectrum is voiced up
to the highest frequency voiced band. Alternatively, the total number of
voiced bands obtained in MBE mixed-voicing can be counted and used to
set the same number of low frequency bands to voiced. In this case, some
high-frequency voiced bands will be swapped with low-frequency unvoiced
bands. Although these methods may give good quality in the majority of
Search WWH ::
Custom Search