8. Speech perception Flashcards
McGurk effect
Ba and Va sound the same with eyes closed
Only lip reading and mouth shape help us to distinguish them
How do we divide speech (speech segmentation)
Possible word constraint: use knowledge of words to match what we hear to words we know and make sense
Coarticulation: helps predict what is coming next.
MORE coarticulation WITHIN a word than between
Segmentation (syllable based or stress based)
Syllable- recognise familiar, clear syllables
Stress- emphasise certain syllables
Categorical perception
Ambiguous phonemes are categorised as one or the other with an abrupt boundary
Phoneme categories vary between languages so much so that some do not sound different to us while they are distinct categories for others
Babies categorical perception
Babies can discriminate between phoneme categories of ALL language until 10-72 months old and then only their native language ones
Adults can only hear their own language
Mattys et al 2005
Hierarchical approach to segmentation
Combines all cues together
Tier 1- lexical. OPTIMAL
Semantic context based on lexical knowledge
SUBLEXICAL
Tier 2- segmental (poor lexical info)
Phonetics and co articulation
Tier 3. Metrical prosody (poor segmental info)
Words stress
Lexical identification shift
Ganong 1980
Varied ambiguous phoneme along continuum from k to g
Inserted this infront of word ending ‘iss’
Found that k was identified sooner as it’s matched to make a real word
(Context affects perceptual changeover point)
Ambiguous initial phonemes are more likely to be assigned a phoneme category when doing so produces a word
Sentence context
Used to help us guess at words or used as a segmentation cue
Eg. Lovers are meant to ADORE each other
The hallway leads to A DOOR at the end
Context phoneme restoration effect
First phoneme of a word in a sentence replaced by a cough
The word heard changed depending on the context of the sentence
Eg.
The *eel was on the axle
The *eel was on the shoe
The *eel was on the orange
Sentence context has an INDIRECT effect involving guessing the identity of the missing phoneme
Cohort model of speech perception
3 stages:
- ACCESS STAGE - perceptual representation used to generate a whole list of words
- SELECTION STAGE- 1 item is chosen from the set (narrowed down as more sounds are available and sentence context etc used)
- INTEGRATION STAGE - Semantic properties used to integrate it into the sentence
UNIQUENESS POINT- where there is only one possible word left that it can be
PARALLEL and interactive processing. Phoneme and lexical and semantics used at the same time to get to uniqueness point
More emphasis on bottom up processing
Isolation point
Point in a word where listeners identify it correctly
Usually earlier than uniqueness point but may be wrong
Lexical access
Point at which all information about a word (phonological, semantic, syntactic) become available following recognition
Fluent restorations
When listening to a sentence with an error. Asked to repeat it back as quickly as possible and 50% of the time the error is corrected automatically
Mispronunciations
Participants are more sensitive to changes at the beginning of a word
TRACE model
Interaction of bottom up and top down processing
Bottom up activation - from acoustics > phoneme > word
Top down activation- word> phoneme > acoustic feature
There are inhibitory connections between units within each level, once activated it inhibits competitors
EMPHASISES TOP DOWN processing
Lexical context can directly effect acoustic perceptual processing
Evidence for top down processing
Lexical identification shift
Word superiority: phonemes detected faster when presented in the context of words rather than non words
Frequency effect: faster identification of common words
