ENT meeting workshop

1
ENT meeting / workshop
Voice Quality Systems

• Analyse og registrering af stemmelidelser Odense
• Friday 6th, Saturday 7th October 2006.Peter
  Pabon
• Pabon_at_let.uu.nl // Pabon_at_koncon.nl
• Utrecht University / Royal Conservatory the Hague
• Voice Quality Systems
• Alphatron Medical Systems BV
• KEBO MED

2
General effects of voice pathology on the
phonetogram

• Total reduction of the size, both dynamics and
  tonal range
• Degrading of the falsetto register.
• More irregular contour, deformation, local
  protrusions in the oval.
• Irregular vibration, increase of additional
  noise.
• Break down of control, the ability to hit a tone,
  unpredictable outcome.
• Breaking up of the oval in small areas, often
  vowel dependent.

3
Aspecific effects, changes that can go two sides

• Upper contour slopes steeper up or gets flatter.
• Discongruence, oval gets more slender or
  thicker.
• Incomplete glottal closure or hyper adducted, too
  powerfull closure.
• Shift/translation to a lower or higher tonal
  range.

4
Interpretation speech area
5
Models / objective parameters,what is normal for
a speaking voice?

• Helbert Damstés rule of thumb, centre first
  octave (6 semitones from lowest pitch).
• Peter Pabon SPL norm 70 dB (many handbooks
  specify 80 dB), gt 75 too loud, lt 65 too soft.
• Rosemary Orr (05), Renée Speyer (dec.03) speech
  position is point of reference, no numerical
  data.
• Sound level seems more important than Fo, but
  no-one is measuring it as most tools dont
  fascilitate this.

6
Speaking voice within total area, some examples
of normal voices
4 ST
7 ST
14 ST
7 ST
8 ST
10 ST
5 ST
8 ST
7 ST
7
Interpretation shapeand contour
8
Reproducability contour measurement higher with
max ( 3 dB). then with min (3..6 dB).
The line is also smoother, more straight, does
not improve with longer integration times, no
shifts or up and down dancing during recording.
More uncertain, jagged, more difficult to pick
one value, instable but not less important!
9
Which contour / shape parameters are important?

• Minimum SPL (Heylen, Speyer, Wuyts-DSI, and many
  others).
• Parameters that rate the phonetogram contour
  around the speech area (Orr, Speyer, Heylen)
• Highest tone, tone range, Intensity range
  (Heylen, normal versus group noduli to test
  relevance)
• Sloping bottom contour (Heylen, noduli) Soft and
  high in falsetto, falsetto is the first area that
  is degrading, high jitter

10
Speyer, multidimensional analysis to factors that
rate therapy effects finds two factors

• Changes in lowest frequency and the area under
  the mean speaking fundamental.
• Changes in minimal intensity together with size
  of speech area
• Renée Speyer, George Wieneke, Ida van
  Wijck-Warnaar Philippe H Dejonckere, Effects of
  Voice Therapy on the Voice-Range Profiles of
  Dysphonic patients, J of Voice, Vol17, no 4, 2003.

11
Comment authors

• Logopeds that cooperated were focussed on
  improvement of the speaking voice. This could
  explain why the largest changes were found in the
  speech area.
• remark speech therapy works!

12
Criterion area gt 90 dB SPL

• Old but very efficient rule of thumb. An
  observation by Wolfgang Seidner, but still very
  efficient.
• Patient if the voice can efficiently be raised
  above background noise level.
• Singer that part of the range that will be heard
  when operating solo on a stage.

13
Normphonetogram untrained male voices n12
14
Normphonetogram, untrained females n20
15
Normphonetogram, children bg, age 9, n12
16
InterpretationVoice Quality Parameters
17
Why Adding Parameters?/Applications

• Measure acoustic voice quality ( Reproduce!)
• Evaluate voice therapy
• Voice Classification
• Locate Registers, (boundaries techniques)
• Study mixing techniques
• Use the direct feedback for training
• Use the Phonetogram as a look-up table to
  Synthesize dynamic voice quality variation.
• Do research on voice production

18
Additional acoustic voice quality parameters that
are in Voice Profiler

• Jitter
• Shimmer
• Crest factor
• Density

19
Color representation of parameters

• The same gray scale or color gradient
  (red-yellow-green) is used for each of the
  parameters, there is no specific color scheme
  reserved for a specific parameter.

20
Voice Quality

• Voice quality depends on one or more acoustical
  features and is therefore measurable.

Warning There is no one-to-one correspondence
between a physical parameter and a perceptual
feature.
21
Addition of acoustical voice quality parameters

• Jitter Duration deviations from
  period-to-period and Shimmer Amplitude
  deviations from period-to-period(i)regularity,
  (a)peridicity or perturbation of the fundamental
  vibration.Roughness, breathiness of the voice.
• Log-scale display of short term SD

High perturbation values are red, less
perturbation yellow, very periodic green.
22
Example Voice Patient, nodules and oedema, Female
23
Normal pattern jitter

• Lowest jitter (lt 0.5) loud phonation,
  modal/chest.
• High jitter (5..1) at the threshold of
  phonation
• Extreme high jitter (gt5 over red!) vocal fry
  and with laryngectomees (lower jitter threshold
  to record).
• Each voice, normal or pathologic shows a decrease
  in jitter (red gt yellow gt green) with increasing
  level.
• With pathologic voices this decrease sets in
  later.

24
What perceptive feature do we measure with jitter?

• Not one specific!
• It is only a coarse/rough physical parameter.
• Low Fo roughness
• High Fo noise, breathiness
• Also moments of fast pitch movements
• Shimmer does comparably, more behaves as a
  turbulence measure.

25
Voice Quality, Jitter
normal
A lot of noise when soft and high in falsetto
Noise over the complete falsetto, and also with
loud voice in modal register
26
Which jitter to take, meaning?

• What to do with the concept of rating voice
  quality with one jitter value?
• It seems that if a vibration stands, if the motor
  runs, it will also be more or less periodic. The
  irregularities come with the transitions (more
  evident in speech).
• Irregularities have a incidental character, come
  in short convoys (too little mass) get
  overshadowed.

27
Interpreting Jitter results

• Depends where you are in the phonetogram.
• High and soft -gt noise/insufficient closure.
• Speech area different opinions on the
  discriminative valueNumerical values often
  incomparable to those from other systems

28
Can we control the Jitter level?

• No, not consciously at will.
• Yes, by holding a steady tone for a long time
  period (no movement whatsoever).
• Yes, indirect by decreasing the area with
  incomplete glottal closure using visual feedback
  option in VP.
• Jitter measurement with the speaking voice
  meaningless as too much things vary at the same
  time!

29
Crest Factor spectral slope
30
Crest Factor as Spectral Slope Parameter

• Measures the peak level in relation to the mean
  energy level per period.

Pure Sine Wave, Steep slope 3 dB green, 7 dB,
yellow, Many harmonics, flat slope, 12 dB, red.
Not so exact, but efficient estimator of
spectral slope
31
Van sine to pulse and back
32
Crest Factor pattern, trained singer, female
33
Normal pattern crest factor

• Highest Crest Factor values loud phonations
  Modal/Chest Register, low Fo.
• Extreme peaking Crest Factor values pressed
  phonations.
• Low Crest Factor values (ca. 4 dB) around
  threshold of phonation in modal register.
• Lowest Crest Factor Values (approaching 3 dB)
  at the threshold of phonation in Falsetto
  register.
• Sharp change (break) ltgt gradual change (mixing)
  loud phonations in the register transition
  zone.

34
Example 1With the crest factor it is possible to
recognize the area that is covered by chest/modal
register.
35
Example 2Crest Factor parameter pattern, voice
patient, femaleno area with strong HF-energy (no
forceful closure) red colour missing
36
Crest factor extreme values

• Extremes with loud male voice gt7 dB
• Lowest maximum values with child voice ca. 5 dB
• Female voice extreme values are somewhere in
  between ca. 6 dB.

37
Affected by

• F1-Fo distance-Mouth opening-Vowel-Fo
• Open-time in the glottal cycle-Pressed
  phonation-Register changes
• Resonance

38
mouth closure damps
39
If it is so imprecise, if so many factors affect
the parameter why keep it?

• Simple to measure, much experimental data.
• Proved its distinctive value in recognition.
• Any other parameter will suffer from the same
  drawbacks.

40
What about other parameters?

• HNR / Cepstrum peak / just variations on jitter
  measurement.
• Still there seems no good breathiness measure and
  research seems to dry out. Spectral models are
  incomplete.
• Difficult to define new parameters that work
  are robust over the complete Fo/SPL range.
• VF-contact only possible with Lx signal tried!
• LF (CQ) and HF spectral slope in spectral ext.
• Singing formant, spectral envalope in spectral
  ext.

41
End part II
42
With louder phonation the voice follows an upward
trace
Physiological ideal trace
43
Mechanism
Physiological principle the sound level
increases with the fundamental frequency and vice
versa.

• A larger vocal fold vibration amplitude also
  means a larger average tension in the vibrating
  vocal folds and thus a higher pitch (passive).
• More tension needed to compensate larger forces
  from higher air pressure (active).

44
A comparable mechanism controls the shape in the
phonetogram of

• Bottom contour/threshold of phonation
• Upper contour, max. phonation level
• Register boundaries

45
Maximum shifts with louder spreech along a
parallel trajectory
Calling voice?
46
Shape changes with pathologies

• In general voice pathology will lead to a
  reduction of the tonal range and the dynamic
  range, a decrease of the total area.
• De contour gets more jagged, with sharp edges up
  and down. Large holes start to show, sometimes
  the phonetogram breaks up in pieces, islands of
  sustained vibration. This makes it more difficult
  to apply general shape parameters in case of
  pathologies.

47
As multiple factors effect the shape, it is
impossible to link one specific shape (change) to
one specific pathology or type of vocal use /
abuse.

• The voice as a musical instrument The more
  physical differences occur (for instance due to
  pathologies), the more different the control
  settings and the more different the resulting
  phonetogram shape.
• Speyer (dec. 2003) The effect of voice therapy
  on the VRP is multidimensional, many different
  changes in VRP are possible

48
Typical shape parameters
49
(No Transcript)