Pitch Estimation and Voiced–Unvoiced Classification of Speech - Digital Speech: Coding for Low Bit Rate Communication Systems - page 162

30

Male

25

Female

20

15

10

5

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SS-SA weighting factor ( ) β

Figure 6.11 Analysis of the effect of the SS-SA weighting factor β in terms of the
pitch error rate. Here, the formant weighting factor γ is 0.9

obtained when β 0 . 3forthemale

speech, which means that the optimum β differs slightly depending on
the pitch period of the signal. Higher performance can be achieved by
weighting the SA more during shorter pitch period speech and less during
longer pitch period speech.

= 0 . 1 for the female speech and β =

Examples of pitch contours of the various PDAs are illustrated in Figures
6.12 and 6.13 in which the rectangular window is applied to the TA and STA.
It shows that pitch errors in strongly-voiced regions are reduced considerably
by the combination of time and frequency domain PDAs. Most of the errors
were caused at speech onset and offset regions where irregular pitch pulse
sequences are present.

6.2.4 Pre-andPost-processingTechniques
In addition to the main pitch determination processes described previously,
there are several important pre- and post-processing techniques which can
significantly improve the pitch determination performance. These techniques
supplement the PDAs and are used before or after the pitch determination
process. Hence, they are usually called preprocessing or post-processing
stages.

Spectrum Flattening
Although the pitch of a voiced speech segment can be directly estimated
from the original speech, the first formant frequency may affect the accuracy