Speech Science XII

1
Speech Science XII
Version 2007-8

• Speech Perception
• (acoustic cues)

2
Topics

• ? Psychoacoustics
• ? Psychophonetics acoustic cues
• Reading BHR, chap. 6, 184-203 (5th
  ed.) chaps. 9/10, 201 ff.
• P.-M., 3.2.2., first part. pp. 158-171
  (2nd ed.) 149-162 (1st ed.)

3
Psychoacoustics 1

• Psychoacoustics investigates the relationship
  between basic (acoustic) signal properties and
  basic auditory impressions
• - How loud something sounds.
• - How high- or low-pitched something sounds.
• - How long somethings sounds.
• - What the timbre (quality) of a sound is.
• The questions asked are
• - Can the signal be heard? (signal strength)
• - Can differences between signals be heard?
  (for all signal properties)

4
Psychoacoustics 2

• Important Psychoacoustics relates the objective,
  measurable signal to subjective impressions.
• These are two different worlds
• The simplest model of psychoacoustic
  perceptionwould be a linear relationship
• - A change in a signal parameter always has an
  equivalent change in the auditory impression.
• This not the case
• (which makes psychoacoustics very complex .)
• Some of the non-linearity has direct
  implications for phonetic understanding..

5
A non-linear relationship Loudness
6
The reason for non-linear loudness
Resonance characteristics of the outer ear
7
Non-linearity above threshold
8
Also, sounds mask one another
If noise is present, a tone has to be stronger to
be heard(it has a higher audibility threshold).
Intensity of pure tone (masked) stimuls (dB)
The closer the tone is in frequency tothe centre
frequencyof the noise, the stronger it has to
be to be heard!
9
Critical Bands (Barks Erbs)
Wide-band noise witha gap still masks a tonein
the middle of the gap
until the gap reachesa critical width.
Then the signal is heardat the same threshold
asif there were no noise.
The noise no longer interferes with the part of
the hearingmechanism dealing with the tone.
These critical bands arenarrow at low and
broader at higher frequencies.
10
Non-linearity of loudness with duration

• Above approx 300 ms (exact duration not certain)
  the perceived loudness of a sound is determined
  by signal strength (and frequency) independent of
  its duration.
• Below this duration, a shorter sound is heard as
  less loud than a longer sound of equal intensity.
• I.e., it is as if the energy is integrated over
  time, so that a shorter sound has less energy
  than a longer one.
• Phonetic importance? Short (unstressed)
  syllables are perceptually less prominent than
  longer (stressed) syllables.

11
Psychophonetics

• Used here as a term to parallel
  psychoacoustics. In our definition,
  psychophonetics is the study of the relationship
  between the acoustic speech signal and functional
  aspects of speech e.g., speech sounds,
  (stressed/unstressed) syllables, tonal accents,
  junctural phenomena etc.
• The experimental procedure typically requires
  changing the analytic properties of the acoustic
  speech signal in a controlled manner and
  recording the perceptual effect.
• The properties changed are those of acoustic
  analysis duration, intensity, fundamental
  frequency and spectral structure.

12
Acoustic Cues

• This term was coined in the 1950s, when
  synthesis and manipulation of the acoustic speech
  signal was starting. (Origin Haskins
  Laboratories, NJ, USA)
• The cues are those acoustic properties that
  can be shown to affect the perception of a speech
  sound.(so we have acoustic cues for vowels and
  consonants, and within these categories
  fore.g. voicing, manner, place of articulation
  in consonants, degree of opening, place,
  rounding etc. in vowels )

13
Acoustic cues vowels 1

• Cues Formants 1 and 2 (to a first
  approximation)

. and the evidence from formant synthesis
14
Acoustic cues - vowels 2

• While monophthongs have a steady state formant
  structure, diphthongs e.g. aI, aU, ?I and
  (vowel glide) approximants e.g. j, w, ?
  have changing formants as a cue to their
  identity.

aI, aU, ?I have a more or less fixed formant
pattern, determined by the identity two vocalic
elements which define them.
j, w, ? have a defined starting point, but
the degree of formant change is determined by the
following vowel. The starting point has a
(slightly more damped) formant structure similar
to the related vowel j ? i w ? u
? ? y (see acoustics slides)
15
Acoustic cues plosives

• Plosives have a temporally complex set of
  acoustic cues resulting from (i) the closing
  movement, (ii) the closure phase and the (iii)
  release of the closure.

The closure is a period with no energy
(voiceless stops) or a weak low frequency
periodic signal (voicing in the closure). This
introduces a perceptible interruption.
The release burst is the result of turbulence
due to the escaping air from the increased
intra-oral pressure built up during the closure.
This may be relatively weak (in voiced stops) or
strong (in voiceless stops).The different
spectral properties of the burst noise signal the
different places of articulation.
16
Release bursts and vowel quality
17
Vowel formant transitions as consonant cues

• Formant transitions (changing formant values in
  the vowel preceding and following the stop
  consonant) reflect the articulator movement
  towards and away from the closure. The F2
  transition is a cue to the consonantal place of
  articulation F1 just signals the opening and
  closing movement.

The place of the stop determines the F2 formant
value from which or towards which the transition
moves (called the locus). But the actual shape of
the transition is determined by the vowel (as it
is with vowel glides).
18
Locus frequencies e.g. d
19
What sort of transitions for which place?

• The previous slide showed that the locus for
  d (and logically for t, n, l, s, z) is
  fairly constant. The value (for the average adult
  male vocal tract) is about 1800 Hz.

For labial consonants, the vowel can be formed
independent of the consonant closure (the tongue
is free to move). Both F2 and F1 therefore just
reflect the opening and closing of the jaw and
lips. The locus is therefore always low.
For velar consonants, the consonant closure is
very dependent on the vowel (both use the tongue
dorsum).The locus is higher than for alveolars
both for front and back vowels, but for back
vowels it is lower than for front vowels. F2 and
F3 transitions often converge with velars.
20
(No Transcript)
21
The importance of timing as a cue to the
voicing distinction
The temporal differencesshown here signal
thedifference between weakand strong
plosives,whether there is closurevoicing
present or not.It is often claimed that
thedistinction fortis-lenis is better than
voiced-voiceless
22
Acoustic cues - fricatives

• Fricative identity is determined by the
  spectral distribution of the energy (see also
  acoustics slides).

D
T
f
v
S
Z
z
s
23
Summary of cues - Manner
24
Summary of cues - Place
25
voice bar
Summary of cuesFortis-lenis