Voice dynamics and spectral slope

1
Voice dynamics and spectral slope
Page 1
Peter Pabon, Voice Quality Systems / Royal
Conservatory Den Haag / University Utrecht
Introduction There is a plain relation between
the sound level and the spectral slope, the
louder the voice, the more high frequency
harmonics are found in its spectrum. With
increased vocal effort, the excitation by vocal
closure will be more effective, that is more
rapid and more powerfull. The closure and thus
the spectral slope also depends on Fo and the
register. Also the low frequency curvature of the
spectral envelope is varied by the relative
duration of the closed phase. Phonetograms are
presented where several spectral slope measures
are displayed as a function of the sound level
and fundamental frequency.
We did not start at zero, we have an idea what to
expect
Doubts about the crest factor as a spectral slope
measure!
Actually the aim of this study is twofold (1)
to investigate the dynamic behaviour of the
spectral slope within the phonetogram, and (2)
to investigate how in a practice the crest factor
behaves in comparison to an official spectral
slope measure
Method Slope parameters are measured using a
pitch period synchronous spectrum analysis
The following four parameters are selected to be
displayed (note the same green..yellow..red
color scale is used for all parameters !)
A Crest factor per period, range 3 dB
(green)...8 dB (red) expected to behave as
a spectral slope parameter B Relative level
band 4xFo, range -40 dB (green)0 dB (red) as
a HF, closure related spectral slope
parameter C Level ratio 1xFo / 2xFo, range -25
dB (green)15 dB (red) as a LF, OQ/CQ related
spectral slope parameter D Relative Level 1xFo
band, range 0 dB (green)-10 dB (red) chosen
because its dynamic behaviour puzzles me
2
Peter Pabon, Page 2
Voice dynamics and spectral slope
Results
Conclusions / Findings
(1) A sudden raise of HF spectral slope, as
indicated in the 4xFo band (parameter B) is
generally accompanied by a first upward going LF
slope as indicated by the level ratio of the
1xFo/2xFo bands (H1/H2, parameter C). So,
forcefull closure and an elongated closed phase
go hand in hand in modal register (red area in B
and C). (2) The pattern of the crest factor (A)
is comparable to that of the 4xF0 band (B), the
official slope measure.The limiting effect of F1
on the cresting of the waveform, as predicted by
the source filter model is not observed. (3) When
Fo goes up, there is a distinct jump where the
relative level of the Fo (parameter D) flips from
overruled to not overruled. This marks the
transition of modal to falsetto register. The
1xFo/2xFo band ratio (H1/H2, parameter C) also
shows this jump. At this moment it is unclear
while the quality jump observed in the first two
parameters A B appears at a higher Fo.
Voice Quality Systems