Pitch Estimation and Voiced–Unvoiced Classification of Speech - Digital Speech: Coding for Low Bit Rate Communication Systems - page 175

Figure 6.22 Original speech waveform and the corresponding pitch similarity plot
with a possible voicing threshold of 0.5 (shown by the dashed line)

which has a value between 0 and 1, indicating no similarity and 100 %
similarity, respectively. Time plots of typical voiced and unvoiced speech
against pitch similarity are shown in Figure 6.22. As can be seen from the
figure the voiced parts of speech clearly have higher pitch similarity than
the unvoiced parts. This is expected since two adjacent unvoiced speech
segments do not possess noticeable similarities.

Peakiness of Speech
Periodic or voiced speech contains regular pulses which do not appear in
unvoiced speech. This feature is described as peakiness of speech and it can
be used to identify voiced speech when it has a relatively high value. In order
to enhance the peakiness, the LPC residual can be used to compute its value.

N

1
N

r 2 (i)

i = 1

Pk = (6.43)

N

1
N

1 |

r(i) |

i =