Fichas
Speech Perception
Fichas del conjunto (33)
- 1How can you interpret a spectrogram?Formants are dark stripes, which show lots of energy at a particular frequency. This is a vowel sound (open vocal tract -> vocal chords vibrate more)
Formant transitions are the dark, diagonal lines which often prelude a formant. They show a high amplitude but changing frequency, and are characteristics of consonants. Lighter parts of the spectrogram are likely consonants (closed vocal tract) - 2What is segmentation?Process of IDing beginnings and ends of words in a continuous sound. Note that we link words together, they're not fully 'segmented'.
- 3What are constraints on segmentation?Top-down restraints, or our knowledge of words, impacts segmentation. We can recognise particular sounds as being a possible word, which leads us to segment them from the rest of the signal.
For example: pps recognised the word 'dull' in the pseudo-word 'dullfep' because 'fep' sounds like it could be a word, and thus it was separated out from the signal, allowing ID of 'dull'. However, 'dull-f' wasn't recognised as containing a word, because just the phoneme 'f' isn't a possible word, so it wasn't separated out. - 4Describe coarticulation.Depending on the context of a phoneme, you perceive the phoneme differently. Effect of needing to prepare for the next phoneme. This is reflected in spectrograms, where formant transitions for the same vowel look different depending on their following consonant.
- 5How are people able to understand what different speakers are saying?In perceptual learning, the person actively learns to interpret the new way of speaking, but this isn't always used
If you expect to hear something unfamiliar, you put extra effort into interpreting it.
For example, when pps were told to expect two different voices but actually heard the same one twice, they had more trouble listening, because they were devoting cog resources into identifying and coping with two diff voices. - 6How do experiments simulate adverse listening conditions?In energetiic masking, the pps hears distracting sounds, which make the words less intelligible by affecting bottom-up processing.
In informational masking, the pps are given distracting actions, which increases their cognitive load, making speech perception harder. - 7Describe the study which investigated categorical perception of phonemes in adults.This study artificially altered the VOT, or voice onset time (the time btwn the the vocal tract closing, and the start of vocal cord vibration), of a syllable to make it sound like 'da' or 'ta'.
There was very little division of pps in perceived phoneme. In general, pps were all agreed as to what phoneme it was. There was a clear cut-off point, or categorical phoneme boundary (a learned categorical distinction btwn phonemes).
In a follow-up study VOT was varied btwn two sounds by 25 ms, and pps said if sounds were the same. If the difference included the boundary, they perceived the difference. Otherwise, thought they were the same. - 8Describe the study which investigated categorical perception of phonemes in newborns.1-month old infants hear sounds w/ varied VOT (changes in 20 ms). Change doesn't include boundary -> sucking rate decreases (becoming bored). Does include boundary -> sucking increases (they detected a change
- 9Describe the study which investigated categorical perception of phonemes in regards to language.Child is taught to look at a speaker when they hear a new sound. Hears speech sounds that exist in Hindi, not English -> could detect Hindi phonemes. Adults (Hindi & English) could only detect phonemes of own language. This ability to detect all phonemes is lost as it's a waste of resources once language is learned
- 10What's the difference between music and speech?There's overlap in the neural activation caused by the speech & speech sounds, but the pattern of activation in these overlapping areas varies.
Impairment can be isolated to only one. - 11What are the processing stages of speech perception?Decoding: selecting what to listen to by extracting the phonemes or syllables of the speech signal, forming abstract representation
Segmentation of sounds: IDed either by syllables or phonemes
Recognition/word ID (40,000 words but 35 phonemes, many sound alike)
Interpretation: constructing coherent meaning using words & order
Integration: integrate w/ preceding speech to understand message - 12What are the general types of cues used for segmentation?Stress: in English, this is typically on the first syllable. In a study, long strings of words where 2nd syallble is stressed -> misinterpretation
Coarticulation: the way phonemes change depending on context, helps anticipate next phoneme. More coarticulation within words than between them -> helps segregate words. Segments or syllables at the beginning and end of words tend to be longer -> helps segregate words
Semantic context: in a study, pps either heard 'a door' or 'adore' in diff contexts. Depending on context (as they had no acoustic cues), reported whether the sound was one word or two - 13Mattys et al suggested a hierarchical organisation of cues for segmentation. What is this theory, and its support?Tier 1 = Lexical (syntax, word knowledge, semantics)
Tier 2 = Segmental (coarticulation, allophony [the way phonemes change depending on the word, not as result of preparation for other sounds])
Tier 3 = Metrical prosody (word stress)
In no noise conditions, lexical cues are more helpful than stress. In noise condition, vice versa.
Coarticulation > stress in segmentation. Harder to use T1 & 2 cues -> stress = more helpful. - 14What are transitional probabilities between syllables?The likelihood that another syllable will follow a given syllable. E.g. -tion has a low transitional probability.
- 15Describe the McGurk effect.People use visual info, even to the point of impairing speech perception.
E.g. see people say 'ba' with audio of 'ga' -> hear 'da', a mix of the two.
This is stronger when the word is placed in a 'semantically congruent' sentence, shows that top-down processes are involved
Even if the visual & auditory inputs aren't synced (and pps area are aware of them being out of sync) effect still exists. Suggests the effect isn't automatic. - 16Describe the two accounts for the effect of context.The interactionist account suggests that context influences early-stage processing and word perception.
The autonomous account suggests it influences late-stage processing and integration. - 17What support is there for the interactionist account of context processing?1: Pps see 4 visuals and hear sentences. If target word sounded similar to name of visual, would fixate on visual, UNLESS context (preceding verb) made visual unlikely in sentence. Fixation happens very quickly, suggests early processing.
2: Pps hear clear/degraded speech + transcription (context)/nonsense letter strings. Rated clarity higher if given transcription. Context activated primary auditory cortex -> influences early stage processing. Hear degraded speech, THEN see a transcription -> doesn't influence rating of clarity. - 18Describe the phonemic restoration effect.Effect of context on phoneme perception.
Phoneme replaced by cough -> heard a phoneme that made the full word make sense in context (wheel, heel, meal, peel)
Happens more with real words than non-words, thus it requires lexical info. - 19What causes the phonemic restoration effect?Two mechanisms:
Sensory repair: successful phonemic recognition leads to the suppression of the auditory cortex, suppressing processing of the the interrupting noise. Thus it's an unconscious process.
Subjective continuity: top-down expectations cause people to perceive the missing phoneme - 20Describe the Ganong effect.Aka the lexical identification shift.
Hear a word w/ an ambiguous initial phoneme -> more likely to hear a phoneme that makes a word
Seen to cause increase activity in areas associated w/ auditory perceptual processing (supports interactionist account of context effects) and in areas associated w/ executive processes (suggests post-perceptual decision is involved, supports autonomous effects) - 21What evidence is there for the autonomous explanation of context effects?In clear speech, context doesn't influence activation of primary motor cortex. This suggests that context takes time to influence processing, as in clear speech, processing is fast, so context isn't considered.
- 22What's the effect of orthography on speech perception?Lexical decision task w/ auditory stimulus. If phonogy & orthography are consistent (e.g. park), quicker to process (e.g. than yacht)
In/consistent show diff brain activation patterns 330 ms after hearing, suggests orthography influences lexical access
Pps decide whether a word ends in a specific syllable. In/consistent -> difference in activation. Supports influence of orthography BEFORE lexical access (just processing the word, not in finding its meaning) - 23Describe motor theory.We mimic a speaker's articulatory movements. This gives less variable & more consistent info about the words than the speech signal does.
Language processing seen to use action-perception circuits, which include various auditory areas, as well as pre-motor and motor areas.
May involve mirror neuron system (some neurons are activated when performing or watching action) - 24What evidence is there to support motor theory?Language processing activates auditory areas, premotor areas, and motor areas.
Transcranial magnetic stimulation to the area of the motor cortex associated w/ lip representations impaired auditory processing of changes in speech sounds. - 25What effect does noisy conditions have on the influence of motor areas on speech perception?In no-noise condition, application of transcranial magnetic stimulation to eh motor areas for the tongue and lips had no effect. In noisy conditions, if stimulating areas relevant to the speech sounds being heard, consonant detection was faster.
Therefore, motor theory may only be relevant in adverse conditions. - 26What evidence contradicts motor theory?Brain-damaged patients with aphasia to the Broca's area have severe issues with speech production, but are still able to perceive speech well.
Damage to the mirror neuron system still yields good speech comprehension.
Anaesthesia of left hemisphere (motor speech system is dominant, but not auditory speech system) -> unable to speak, but could comprehend speech.
Transcranial magnetic stimulation to the premotor cortex impacted a phonemic task, but had no impact on a semantic task. Stimulation to posterior superior temporal gyrus (used in auditory & speech processing) impacted both. Shows motor areas is less influential that auditory & speech processing areas - 27Describe the TRACE model of speech perception.Connectionist network, with 3 levels: features (voice, tone), phonemes, and words. All levels are connected, with bidirectional facilitatory connections between levels, and inhibitory connections within levels. The influence of a node depends on its own activation. All information is available.
Bottom-up processing: from feature -> phoneme -> word. Top-down processing: opposite.
Predicts a word superiority effect & frequency effect. - 28What support is there for the TRACE model?Research has confirmed the word superiority effect, as ppl detect phonemes better in words than non-words. This effect was greater when most of the stimuli were words.
The model explains the Ganong effect, as top-down processing would cause activation in phonemes that would make a word, even if the phoneme isn't present.
Pps' perception of an amibuous phoneme is influenced by recent activation of possible phonemes. Shows influence of prior activation.
Explains discrimination boundary, as it becomes sharper via mutual inhibition of phonemes.
Clear word-frequency effect 340 ms from word onset. - 29What evidence contradicts the TRACE model?Too much importance on top-down processes. Able to detect incorrect target phonemes in very word-like non-words. Top-down activation from word level should should impair (e.g. vocabuTary). Also, didn't show preference for perceiving a terminal phoneme that would create word when ambiguous, except in degraded stimulus situation.
Doesn't consider effect or othography. Words that are orthographically inconsistent are slower to identify via audio.
Mispronunciations have a greater effect than top-down processes suggest.
Syllable duration also helpful. Pps able to determine whether recording would say 'cap' or 'captain' after only hearing 'cap'. - 30Describe the Cohort model of speech perception.Potential words (according to what you've heard so far) are in cohort. Competition btwn these, eliminated as sound/context continues. Uniqueness point: point where only one word is possible.
Later updated so membership depends on activation level, and based on SIMILAR-sounding phonemes, not necessarily the same. Less black and white. Distorted first phoneme has been seen to take longer to identify.
Processing stages:
Access stage: cohort activated
Selection stage: word chosen
Integration stage: semantic & syntactic properties are integrated
Original: context affects selection
Update: context affects integration - 31What evidence is there to support the Cohort model?N400 (measures word processing) appears earlier when uniqueness point is earlier. Suggests its importance.
Words that are semantically similar to possible completions of an initial syllable (cap- -> captain -> ship) are primed, even if context makes the word (ship) unlikely. Words in initial cohort were primed, regardless of context, showing context affects later stages.
Supporting update: pps hear incomplete words, and fixate on pictures of words that could fit in context & those phonologically similar. So sound similar -> activated, regardless of incorrect context. - 32What evidence contradicts the Cohort model?A study found that uniqueness point was only influential when words were spoken at a slow or intermediate speed, but not at the fast, conversational speed.
Context may affect earlier than uniqueness point. Before hearing uniqueness point, pps show diffs in N400 depending on whether words fits in context. - 33Describe the 5 components of the cognitive neuropsychology three-route framework.5 components: auditory analysis system (extracts phonemes from speech)
Auditory input lexicon: contains info about spoken words, not semantics. Damage -> pure word deafness. Impaired phonemic processing causes inability to perceive verbal words
Semantic system (contains semantics)
Speech output lexicon (contains spoken form of words)
Phoneme response buffer (contains distinct speech sounds)
Información
- Descripción
- Language & Memory Fichas sobre Speech Perception, creado por Jamie Otten el 20/12/2018.
- Creado por
- Jamie Otten
- Visitas
- 0
- Valoración