Digital Signal Processing Reference
In-Depth Information
and the period or cycle jitter
J
c
of
the deviation of the current fundamental period and
the 'ideal' fundamental period
T
0
as obtained by averaging in the analysis interval:
J
c
=
T
0
(
n
)
−
T
0
.
(6.60)
Jitter is known to be particularly high at the beginning and end of a sustained voiced
sound.
In a similar way, shimmer is the deviation of amplitude—usually in dB—from
period to period. A healthy speaker's shimmer is usually between 0.05 and 0.22 dB
[
42
].
To end this section, Fig.
6.2
gives some example plots of speech LLDs as discussed
above.
6.2.2 Music Descriptors
In this section we will deal with LLDs tailored in particular to the analysis of music.
However, many of the previously discussed features are also used for music analysis.
We will first look at basic Pitch Class Profiles (PCP)—in particular by CHROMA-
type features. These are suited for the tonal analysis of music. We will then take a look
at the music theoretic and human perception based variants as were first introduced
6.2.2.1 Pitch Class Profiles
In music theory, 'notes' are characterised and named by their pitch class and their
octave, where an octave is an 'interval' between two notes. An increase by one octave
resembles a doubling of a note's frequency. It is further a special interval: Two notes
played in different octaves sound nearly equal to human listeners and thus share the
same name with different octave number. In western music, the octave interval is
divided into twelve equally sized intervals with the tempered scale. These intervals
are called semi-tones. Their names in western music are shown in Fig.
6.3
that also
visualises the discussed principle.
PCP features are based on the principle of providing the spectral energy per semi-
tone band. They are computed using a DFT with a suitable window length, window
function, and a window overlap—typically around 0.5. Human loudness sensation
can be taken into account, e.g., by applying the A-weighting according to DIN EN
61672-1:2003-10 to the DFT magnitudes. The weighting is given by:
12200
2
f
4
·
1
H
A
(
)
=
·
f
2
·
f
2
9
2
.
f
(6.61)
f
2
6
2
f
2
12200
2
(
+
20
.
)
·
(
+
)
+
107
.
7
2
+
737
.
