Speech Science IX

1
Speech Science IX

• How is articulation organized?

2
Topics

• Same sound, different gestures?
• How do we control our articulation?
• At what level do we control our articulation?-
  individual muscles? - gestures for individual
  sounds?- sequences of gestures for syllables?-
  . for words?
• Reading BHR, Chap. 5, pp. 134-173 ff.
  (Variation, Feedback, Prod.-Models)
• P.-M. 1.4,8. pp. 64-78 (Steuerung)

3
Sound Variants

• A fundamental problem in speech is the
  many-to-one relationship between speech events
  and their causes.
• One acoustic event (speech sound) can result from
  different articulatory configurations
• ( articulatory compensation)
• One articulatory configuration can result from
  different patterns of muscular activity
• ( neuromuscular compensation)

4
Sound Variants (example)
5
Sound Variants (other example)
Another well-known example of articulatory
compensation is the American /r/ (?)
The tongue may be a) turned back
(retroflex) b) bunched
Lip-rounded vowels (like y) can be produced
with strongly rounded, protruded lips, or with
retracted tongue and neutral (or even spread)
lips (with or without a lowered larynx).
6
Free variation vs conditioned variants
Articulatory differences (requiring different
commands to muscles) are not only the
result of having acquired a particular variant.
Sounds occur in context, and the gestures
are different with every different preceding
sound!
This makes the relationship between one
speech sound (in linguistic terms a phoneme)
and the commands to produce it complicated.
7
Context variation Coarticulation
Example Tüte
t u? t
?
When a property of one sound affects the way in
whicha neighbouring sound is produced, we call
the effectcoarticulation. Here lip-rounding in
the initial /t/ of Tooter.
8
Different movements in context
Contextual differences in gestures affect every
part of the articulatory patterns. Here, chin
and tongue-tip interaction.
9
What does this say about speech production?
The motor activity involved in producing
speech sounds is much more complex than the
(relatively simple) phonetic/phonological
categorisation of speech sounds
We have to decide whether there is (can be)
any link between linguistic description and
production models
. It would be unfortunate if we had to say
that the two had nothing to do with each other!
10
But the observations also tell us .
The acoustic (? perceptual) identity of
sounds seems more important than the motor
equivalence
When we learn to articulate, we match our
own production to what we hear.
The acoustic patterns from other speakers are
our only models
Nobody shows us how to move our lips, tongue,
velum or larynx .
11
although
There seems to be an innate ability to
imitate peoples facial expressions
This has been systematically observed in very
young babies, who mimic their mothers
expressions.
So there may be some visual input as well as
the predominant acoustic input to the speech
learning process.
But only a small fraction of the articulatory
activity is visible/observable.
12
Must our normal production processes be related
to processes used in learning?
Theories of speech production do not always
model articulation from a perceptual standpoint.
Linguistic (phonological) models of sound
systems are concerned with the patterns of
sound produced, not with the processes that
are required to produce them (BHR p. 152 f.)
The IPA system is articulatorily orientated.
Distinctive Feature theory (more abstract) can
be articulatory or acoustic.
13
Speech Science X

• Production models

14
How can we model the production process?
We need to explain how we control the
complex articulatory activity in terms of the
different sounds that we can hear.
Whether or not the orientation is perception,
the change from one sound to another has to
be explained.
How do we move the articulators to where
they have to be?
How do we know when they have reached the
correct position so that we can move on to
the next sound?
15
How do we move the articulators to where they
have to be?
How is the movement defined?
- as a specific gesture?
- as a gesture of a specific duration?
- as a target position of the (main)
articulator?
- as (a position related to) an auditory
percept?
- as a spatial configuration?
The target concepts are closely related to
one another
All the ideas take the sound segment (phoneme)
as the unit which is being controlled.
16
How do we know whether thearticulators have
reached their target?
There are a number of feedback channels
- visual feedback? (no use in speech)
- auditory feedback? (too late)
- tactile feedback? (yes, but not enough
- kinaesthetic feedback? (yes)
- proprioceptive feedback? (yes)
Tactile and proprioceptive feedback provide
information about the momentary position.
So we know whether an articulatory action
feels right or not.
17
Types of feedback?
Closed loop reports on a preceding action and
this triggers the next command
18
Types of feedback 2
Open loop allows a sequence of commands to
be carried out
19
Do we use closed loop feedback at all?
Open-loop feedback allows rapid sequences of
commands to be carried out (we dont think
about each syllable, but we still have to
monitor events)
Closed-loop feedback is useful (and necessary)
for units that take long enough for us to
observe them .
and for units that require some conscious
decisions or planning. E.g. at phrase level
(phonologically, the intonation phrase)
20
Evidence for open loop at syllable level
Khozhevnikov Chistovich (KC) from
Leningrad provided the first evidence in den
60s and 70s, showing different degrees of
variance in different size units.
In sentence repetitions of "Tonya topila
banyu", they found that syllable-duration
variance was greater than phrase duration
variance.
and they found that neighbouring syllables
correlated negatively with one another (i.e.,
if one was shorter the next was longer)
21
But what is the basic production unit?
KC looked at syllables in the phrase, but
they didnt take for granted that the syllable
is THE basic unit of articulatory planning.
They deduced it from the patterns of effects
with changes of speech rate. They found
- Rate changes occur between phrases,
- Rate changes do not affect the relationship
between syllables and words,
- But rate changes do affect the relationship
between consonants and vowels.
22
Syllable units but CV together)
KCs results and our lip-rounding observations
indicate that C and V commands seem to be
issued at the same time.
23
Control of gestures towards targets
We have still not explained how the very fast
and muscularly complex gestures are controlled.
Different muscle tension, and even different
muscles are important for one sound in
different contexts.
An automatically feedback-controlled
servo-system would be the engineering solution.
24
Muscle spindles as a servo system 1
25
Muscle spindles as a servo system 2
26
Muscle spindles as a servo system - summary
The muscle spindle is set to the required
target tension.
Fast afferent (feedback) nerve fibres report
on the discrepancy between momentary muscle
tension and target tension.
Fast commands direct to the muscle correct
the discrepancy between momentary muscle
tension and target tension.
target reached, independent of the position
of the articulator prior to movement.
27
Mass spring model of speech production 1
The picture painted so far is simple because
we have only considered one muscle.
As we have discussed before, any one
gesturerequires the coordination of many muscles
.. and any one sound requires the
coordinationof a number of gestures (tongue,
lips, etc.)
The target position for any given sound isthe
product of all the target tensions of
allparticipating agonist and antagonist muscles.
28
Mass spring model of speech production 2
If all the target tensions are set, the
servo-system moves the articulators
automaticallytowards the target position.
This is compared to a mass moving under the
influence of a (damped) spring. The
movementstops when the point of equilibrium is
reached.
Advantage of model It explains undershoot
(a frequently observed reduction of a sound).
How? Another target is defined before the
previous one is reached,
29
Coordinative structures
Disadvantage of the mass spring model
asoriginally conceived
It cannot explain the compensation for
articu-latory disturbances that are observed!
It is therefore assumed (no real proof) that
the mass is made up of sub-structures which
worktogether and compensate for each other.
A well-documented coordinative structure of
this kind is the jaw lips but also the jaw
tongue.
30
Coarticulation vs. Co-production
Coarticulation assumes that features from
one phonemes spread into a neighbouring one.
- There is no explanation for the different
durationsof sound segments (as a function of
stress tempo).
- There is no explanation for how certain
featuresspread. In short, no real link to
production models.
Co-production assumes a fixed (but
unknown)duration for each speech sound. Phonemes
arenot abstract, but concrete articulatory
things.
Different durations are the result of
differing degreesof overlap.
31
Advantage of a co-production view
Stronger (stressed) syllables have less
overlapamong their segments. Unstressed
syllables have more overlap.
- More overlap implies that the command for
thefollowing segment comes before the target of
thepreceding segment is reached.
- This results in a reduced realisation
(shorterand often spectrally less well defined).
But the different phonetic realisation is not
theresult of a different articulatory plan for
the sound.
32
Summary
We have shown that a target-orientated
modelcan be explained physiologically
We have seen that feedback is vital and canbe
used in an automatic servo-system.
We have seen that complex articulatory
patternsa) depend on mutually compensating
musclesynergies, b) can fit into packets of
lower-order commands (macros) which allow more
complex units tobe produced automatically.