Speech Perception

1
Speech Perception

• 4/6/00

2
Acoustic-Perceptual Invariance in Speech

• Perceptual Constancy or Perceptual Invariance
• Perpetual constancy is necessary, however,
  acoustic properties of individual phonetic events
  vary across speakers, across conditions (speaking
  rate, repetition of the same sound)
• Five theories have been proposed to explain
  variance and perception

3
Prototypes or Templates

• Ideal Forms of the acoustic pattern for a
  phoneme
• Incoming stimuli is compared to the ideal and the
  closest match is selected

Store of Templates
Output Decision
Speech Output
Comparator
4
Prototypes or Templates

• Template serves as a reference pattern for
  incoming sound
• Incoming stimuli may be normalized or
  preprocessed to compare with prototype
• Can process intertalker differences (age
  gender)
• Perceptual magnet effect
• Sounds similar to the prototype are identified
  with it
• Perceptual magnet warps the perceptual space,
  so acoustic patterns close to magnet are
  perceived as similar

5
Features

• Features
• Various attributes that compose stimuli are
  identified through abstraction
• Defining attributes of elements to be recognized
• Nonspeech example young girl (female, human,
  young)
• Speech sounds can be analyzed in terms of
  characteristic features
• b bay stop, voiced, bilabial
• Change in features can make the difference in how
  a sound is recognized

6
Features

• Drawbacks features are difficult to describe and
  define
• Features in many systems are composites of many
  features articulatory, acoustic, perceptual
• Speakers analyze sounds in terms of features
• Confusions are feature based p instead of b

7
Higher Order Variables

• Invokes higher order variables as the means for
  perceptual constancy
• Acoustic events are uniquely specified by time
  varying properties
• Listener learns that patterns of energy arise
  from a consistent sound source.
• Problem How are variables identified for complex
  stimuli?

8
Innate Systems

• Invariant cues are identified by innate
  mechanisms
• Simply born with ability to achieve perceptual
  constancy
• Problem How do you account for effects of
  sensory experience?
• Innate mechanisms are the result of evolutionary
  adaptations by which sensory systems are equipped
  to reliably perceive the environment

9
Connectionst Networks

• Parallel distributed processing (PDP)
• Behavior can be modeled with networks that
  involve large numbers of interconnected units
• Modeled from computer theory
• Learn patterns that represent the structure of
  phonemes
• Network learns the structure through various
  adjustments until the input matches the output
• learning language patterns sometimes make the
  same kind of developmental errors committed by
  children

10
Experimental Phenomena in Speech perception

• Categorical Perception
• Laterality Effects
• Adaptation Effects
• Trading Effects

11
Categorical Perception

• Listener categorized stimuli so that the stimuli
  within a category sound alike
• The continuum of a stimuli must vary along some
  parameter (VOT) and has one or more boundaries
  that divide the stimuli into perceptual
  categories- Identification Experiment (label
  stimuli)
• ex. VOT differences with voiced voiceless
  phonemes
• ex. same categorical perception with a color
  spectrum
• Speech categories voice onset time, place of
  articulation, rise time

12
Categorical Perception

• Discrimination Experiment-
• Stimuli within a category be poorly discriminated
  compared to stimuli from different categories
• ex. VOT values of 40-50 ms should be hard to tell
  apart
• Same category- hard to tell apart
• Different category- easy to tell apart
• Categorical perception has been studied using
  VOT-
• VOT stimuli are heard as 1 of 2 categories-
  Voiced (b) or voiceless (p).
• b is chosen with values less than 20 ms
• p is chosen with values higher than 30 ms

13
Categorical Perception

• Categorical perception of VOT for nonspeech
• Tones are distinguished the same as speech VOT
  (perception may be with acoustic stimuli)
• Categorical perception of VOT in infants-
• Show adult like discrimination (heart rate
  deceleration)
• Categorical perception of VOT in Animals-
• Chinchillas can discriminate VOT differences
  (acoustic discrimination based on the mammalian
  auditory system)

14
Responses of b and p to Voice-Onset Stimuli
Phonetic Boundary at 20 ms
15
Laterality Effects

• Right or left ear advantage
• Dichotic perception (delivery of different
  stimuli to different ears)
• The advantage or most heard words will go to one
  ear or the other
• REA (Right ear advantage)- for stops mostly
• LEA (Left ear advantage)- for tones steady
  state vowels
• Some sounds are processed better in one
  hemisphere than the other
• REA is a left hemisphere advantage

16
Adaptation

• Repeated exposure to an extreme member of some
  acoustic continuum causes a shift in a category
  boundary
• Auditory system adapts to repeated stimuli
• Stimulus becomes less effective in evoking the
  phonetic property
• ex. voiceless sound becomes harder to distinguish
  after repetition

17
Trading Relations

• Sounds associated with two or more acoustic cues
• ex. stop consonant (brief silence, burst,
  transition)
• Integration of cues into single phonetic event
  recognizes the different cues as belonging to
  that sound
• Phonetic contrasts Witch and Wish
• Differ in final consonant
• Distinguishing difference is duration of silence
  frication
• The listener evaluates 2 or more cues and makes a
  relative decision based on their properties

18
Trading Relations

• How does this phenomenon take place?
• The cochlea functions as a sensitive detector
  and spectral analyzer of brief transient events

19
Segmentation

• What about distinguishing between words?
• Listener uses a variety of cues-
• rhythmical pattern- stress prosody
• I.e. strong syllable most likely to initiate word
  onset
• Silence or Pausing- often pause between words
• Phonotactics- phoneme arrangement is set
• I.e. velar nasal n as in thing cannot occur
  in word initial position