1. SPEECH PSYC 330: PERCEPTION

2. SOME BASICS Methods of Manipulation PHONATION (air pushed across vocal cords) Airflow Mass and “tuning” of cords Harmonics ARTICULATION (changes in vocal tract – ah; ee) Vocal tract (everything above the larynx) acts as resonator Change shape  change in resonance characteristics (shape increases/decreases energy at different frequencies)  Filter function, peaks in wave called FORMANTS  lowest freq = F1, F2, and so forth

4. ARTICULATION AND SPEECH Vowels TONGUE Up or down Front or back LIPS Degree of rounding EE, AH, OO

6. Consonants PLACE of articulation Bilabial (lips): b,p,m Alveolar (teeth): d,t,n Velar (soft palate): g,k,ng MANNER of articulation Stops: b,d,g,p,t,k Fricatives: s,z,f,v,th,sh Laterals and glides: l,r,w,y Affricates: ch,j Nasals: n,m,ng VOICING Voiced: b,m,z,l,r Not voiced: p,s,ch

7. ADDITIONAL COMPLICATIONS Co-articulation Articulation of one speech sound overlaps with the next, because we talk so fast Adjust production of sounds based upon sounds preceding and following Context effect(say “moody” and “eedoom”  lack of physical invariance in the stimulus – doesn’t both speech perception in practice, but big problem for theory (and for AI) Categorical perception 1.sharp labeling (one OR the other) 2.inability (or difficulty) discriminating within categories 3.discrimination performance predicted by labeling McGurk effect – mismatched auditory and visual input visual “gah”; auditory “bah”  perception “dah”

8. Speech Segmentation Saffran et al (1996) 8 m old infants trained with 2 m stream of artificial language After brief exposure already picking out the “words” DV listening time to presented “words” and “non-words” Concluded that children pick out covariance of sound combinations (statistical likelihood) pri-tee (baby, good, far, nice) bay-bee (girl, boy, good) tee-bay (?)

9. LANGUAGE DEVELOPMENT Can you say 7,777 in Swedish? Not just pronunciation, but hearing too Preference methodology In utero (HR changes) Familiarity effect (own language, own mother, own stories) 6 m Preference for own vowels 12 m Preference for own consonants Huppi & Dubois (2013) brain scans on premature babies (up to 3 mo early – brain not fully developed) Found that they discriminated between male and female voices Found that they discriminated between “ga” and “da” Used same regions of the brain as adults do to make the discriminations

10. SPECIAL TOPIC: EMOTION PERCEPTION IN LANGUAGE Laukka (2005) Categorical Perception of Vocal Emotion Expression Stimulus Development: Actress says “It is now 11 o’clock” with tones reflecting anger, fear, happiness and sadness Physically “morph” the sounds from one emotion to the next  continuous variation

12. METHOD UGs presented with sequential discrimination task with two tones (ABX) is X = to A or B? All combinations of the morphed sounds differing by 20% were compared Also asked to judge the emotion (in addition to discriminating it)

13. IDENTIFICATION ACCURACY

14. REACTION TIMES: 50% V OTHER

15. DISCRIMINATION: ACROSS CATEGORY V. WITHIN CATEGORY

16. BRAIN STRUCTURES IN SPEECH COMPREHENSION Old school: Broca’s and Wenicke’s areas lateralization effects AI; belt and parabelt  anterior temporal lobe HOW? Rosen et al (2011) How (on what basis) does language (as opposed to other complex sounds) become lateralized? IV = intelligible vs unintelligible sentences of equal auditory complexity (created by manipulating frequency and amplitude changes in auditory signal) DV = brain scan data (PET) Results: intelligible sentences processed in left temporal lobe; equally complex but non intelligible sentences processed bilaterally

17. ATTENTION AND SPEECH "Mechanisms Underlying Selective Neuronal Tracking of Attended Speech at a 'Cocktail Party” Golubic et al, 2013 The Cocktail Party Phenomenon – how do we do it? Direct electrical recording from the brains of epilepsy patients Presented naturalistic stimuli re: “cocktail party” Findings: brain regions in and near to primary auditory cortex respond to both attended and non attended speaker; processing in subsequent paths are selective We use bottom-up processing (temporal/amplitude patterns) to “tune” the selection – selectivity “unfolds” (becomes more prominent) across a sentence

18. Green dots = brain regions that responded to both speakers (attended and ignored) Red dots = brain regions that respond selectively (only to attended speaker)