11 Speech Spectrograms 11 Speech Perception Basic perceptual unit -- phoneme Sounds making up a spoken language Smallest aspect of speech If changed -- changes meaning Ex: fog dog /f/ to /d/ Phonemes Sound -- depends on how articulated Shape of mouth, position of tongue, vocal folds, etc. 11 Speech 11 The Basic Components of Speech Production (Part 1) Humans are capable of producing lots of different speech sounds 5000 languages spoken today, utilizing over 850 different speech sounds Flexibility of vocal tract: Important in speech production 1
11 The Basic Components of Speech Production (Part 2) 11 Phonemes in English Vowels Open vocal tract Vocal Cords Vibrating Change jaw, lip, tongue position Demo 11 Sound from Vocal Folds Harmonic spectrum immediately above vocal cords Shaping of vocal tract accentuates some frequencies, attenuates others (Resonance) 11 Formants and Vowels Position of Formants Unique for each vowel Does not depend on preceding or trailing phoneme Result: Peaks in harmonic spectrum Called formants Numbered: F1, F2, F3, ) 2
11 Formants Can identify vowels based on relative frequency of 1st & 2nd formants Invariant 11 Phonemes in English Consonants Closing or constriction of vocal tract 11 Production of Consonants Consonant production -- 3 main factors Phonation or Voicing Place of articulation Manner of articulation 11 Consonants: Voicing Voiced vs. Unvoiced Vocal cords vibrating? /d/ vs. /t/ Dad vs. Tad /v/ vs. /f/ Van vs. Fan 2 different levels (voiced or unvoiced) 3
11 Consonants Place of Articulation Where is the air stream obstructed? Examples Alveolar -- alveolar ridge /d/ -- dad Labiodental -- teeth and lips /f/ -- fad Bilabial -- two lips /b/ -- bad ~ 9 different 11 Consonants -- Manner of Articulation Manner of Articulation How is air pushed through opening? Examples Stop consonants (plosives) -- air is stopped and released /d/ -- door Fricatives -- air pushed through restriction /s/ -- soar Nasal -- soft palate closes -- air through nose /n/ -- nor (hold nose) ~ 6 different 11 Consonants 11 Phonetic Space Combine to produce specific phoneme Examples /d/ -- voiced, alveolar, stop consonant /f/ -- unvoiced, labiodental, fricative /m/ -- voiced, bilabial, nasal Unvoiced, bilabial, fricative? 4
11 Consonants and Spectrogram Initial Consonants Formant Transitions Ramping up Ramping down Can we use formants to identify consonants? 11 Coarticulation Don t produce phonemes in isolation Leftover shape from last phoneme Preparing for next phoneme dee vs. doe Mouth -- different shape Speech signal different (formant transitions) 11 Coarticulation Physically /b/ are different /d/ are different Perceptually Sound the same In fact eebah and oodah Identical physically Hear as different Mapping physical stimulus to phoneme Use surrounding context Multiple auditory cues 11 Other problems Variability Speaker-to-speaker variability Accents Within-speaker variability Different physical stimuli (speech signal) Nonetheless -- perceive as identical 5
11 Categorical Perception Variability in speech signal Perceive categories of sounds Ignore differences 11 Voice onset time /ba/ vs. /pa/ Example Voice Onset Time -- VOT Time between release of air & voicing Ex: /ba/ vs. /pa/ 11 VOT 11 VOT Demo Manipulate VOT Sounds between prototypical /b/ and /p/ Can t tell difference Hear as /b/ Switch to /p/ Nothing in between Compare with pitch perception Increase frequency Continuous increase in pitch 0 10 20 30 40 50 60 70 80 Web Demo: Ba/Da/Ga Ba/Pa VOT 6
11 Categorical Perception >140 phonemes We become tuned to phonemes of native language Ignore distinctions not used native language Sometimes can detect difference (but ignore) Cat vs. Cat English -- Aspirated vs. non-aspirated /t/ Same phonemic category Other times, cannot perceive difference Rope vs. Lope Japanese -- /r/ and /l/ fall in same phonemic category 11 Infants Initially capable of discriminating all phonemes from all languages 6 months -- start to show initial evidence of categorical perception 12 months -- adult-like discriminations (native language) Movie Clip 11 Why useful? Variability within and between speakers Won t always have perfect phoneme Ignore variability Extract message What is really important 11 Motor Theory How special is speech? Motor theory of speech perception: Special mechanisms just for perceiving speech Perception linked to production Figure out how speaker produced sounds (how we would produce ourselves) Allows us to recover phonemes McGurk Effect 7
11 Problems for motor theory Auditory Theory Nothing special about speech Process speech the same as any other sound Evidence against motor theory: Categorical speech perception: Not limited to speech sounds musical intervals,faces, facial expressions Nonhuman animals can learn to respond to speech signals in similar way to human listeners 8