PowerPoint Presentation

Acoustic and Physiological PhoneticsVowel Production and PerceptionStephen M. TaskoLearning ObjectivesReview source-filter theory and how it relates to vowel productionDistinguish between source spectrum, transfer function and output spectrum.Calculate formant/resonant frequencies of a uniform tube based on its physical dimensions.Describe how the area function of an acoustic resonator is determined.Distinguish between and describe relation between area function and transfer function.

Stephen M. TaskoSource Filter Theory

Source(Phonation)Filter(Resonator)Speech(What We Hear)Input SpectrumFrequency ResponseCurve(Transfer Function)Output Spectrum

Stephen M. Tasko33Same Source, Different Filter

Stephen M. Tasko44Stephen M. TaskoFrequency response curve/Transfer FunctionFRC peaks resonant or formant frequencyTube resonators have an infinite number of formantsF1, F2, F3 denotes formants from low to high frequency

F1F2F3F45Stephen M. TaskoVocal tract as a tubeTubes have physical characteristics (shapes)Tubes act as acoustic resonatorsAcoustic resonators have frequency response curves (FRC), also known as transfer functionsTube shape dictates the frequency response curve.6The vocal tract shape during vowel productionCan be (roughly) uniform in shapeThe vocal tract is fairly uniform in its cross-sectional diameter for neutral or central vowel (schwa)Can also be take on non-uniform shapesAre observed for non-neutral vowelsHave a more complex geometryDoes not allow simple calculations of formantsFormant values are derived from the vocal tract area functionStephen M. Tasko7Stephen M. TaskoVocal tract as a tube

Vocal tract: bent tube, closed at one end, with differing Cross-sectional diameter.Straight tube, closed at one end,of differing cross-sectionaldiameterStraight tube, closed at one end,with a uniform cross-sectionaldiameter8Stephen M. TaskoWhat is an area function?Area (cm2)Length along tube (cm)9Stephen M. TaskoArea function of a uniform tubeArea function dictates the frequency response curve for that tube

Area (cm2)Length along tube (cm)10

Vocal Tract Area FunctionStephen M. Tasko11

Vocal Tract Area FunctionStephen M. Tasko12

FRCRelationship between vocal tract shape, the area function and the frequency response curveStephen M. Tasko13Key pointsVocal Tract has a variable shape, thereforeIt is a variable resonatorCan have a variety of area functionsCan generate a variety of frequency response curvesA given area function can lead to one (and only one) frequency response curveA given frequency response curve and arise due to a variety of different area functions

Stephen M. Tasko14Learning ObjectivesDescribe the basic shape of the area function for the four corner vowels.Describe F1-F2 relations for English vowels with specific emphasis of the corner vowelsDraw and recognize (1) wide band spectrograms, (2) spectrum envelopes, and (3) frequency response curves for the corner vowelsDraw and interpret various plots that relate formants values for English vowels.Outline our basic tongue and lip rules for predicting formant shifts from the neutral position.

Stephen M. TaskoVowels: Articulatory Description

Stephen M. Tasko16Vowels: Articulatory DescriptionDegree of lip roundingRoundedUnroundedDegree of tensionTenseLaxStephen M. Tasko17

Neutral ConfigurationVocal Tract Area FunctionFrequency Response CurveArticulatory Configuration/ Vocal Tract ShapeStephen M. Tasko18

Low back vowelVocal Tract Area FunctionFrequency Response CurveArticulatory Configuration/ Vocal Tract ShapeStephen M. Tasko19

High back rounded vowelVocal Tract Area FunctionFrequency Response CurveArticulatory Configuration/ Vocal Tract ShapeStephen M. Tasko20

Frequency Response CurveVocal Tract Area FunctionLow front vowelArticulatory Configuration/ Vocal Tract ShapeStephen M. Tasko21

Relationship between vocal tract shape, the area function and the frequency response curveFrequency Response CurveVocal Tract Area FunctionArticulatory Configuration/ Vocal Tract ShapeStephen M. Tasko22What distinguishes vowels in production and perception?Resonant (formant) FrequencyF1, F2 frequency are particularly importantF3 frequency plays a smaller role

Landmark study: Peterson and Barney (1952)

Stephen M. Tasko23

Vowels: Spectrographic PatternsStephen M. Tasko24Mid Central vowelF1: 500 HzF2: 1500 Hz/i//u/////Gain

frequencyVowels: Frequency Response Curve PatternsStephen M. Tasko25Stephen M. TaskoObservations/i/ & /u/ have a low F1

// & // have high F1

Tongue height ~ F1Tongue height F1 Tongue height F1

/u/ & // have low F2

/i/ & // have high F2

Tongue advancement ~ F2Tongue front F2 Tongue back F2

Learning ObjectivesOutline the key assumptions and parameters of the Stevens & House (SH) articulatory model of vowel production.Describe the acoustic consequences of changing SH model parameters.Provide acoustic explanations for how (1) the SH model parameters influence area function and (2) how these area function changes influence acoustic (i.e. formant values)Compare the shape of the vowel quadrilateral and the F1-F2 plot

Stephen M. TaskoConnecting the dotsHow do articulatory processes map onto acoustic processes?

Stephen M. Tasko283-parameter model (Stevens & House, 1955)Model assumesNo coupling withNasal cavitytrachea & pulmonary system

Stephen M. Tasko293-parameter model (Stevens & House, 1955)Model parametersDistance of major constriction from glottis (d0)Radius of major constriction (r0)Area (A) and length (l) of lip constrictionA/l conductivity index

Stephen M. Tasko303-parameter model (Stevens & House, 1955)

Stephen M. Tasko31Key Goal of StudyEvaluate the effect of systematically changing each of these three vocal tract parameters on F1-F3 frequencyStephen M. Tasko32General Observations

Stephen M. TaskoGeneral Observations

Stephen M. TaskoGeneral Observations

Stephen M. TaskoInterpretation: Double Helmholtz Resonator ModelStephen M. Tasko

Back CavityFrontCavity

MajorConstriction (ro)Back Cavity Volume influences F1Larger volume = lower F1Smaller volume=higher F1

Front Cavity Volume influence F2Larger volume= lower F2Smaller volume=higher F2

Radius of Conduit (r0) influences F1 Larger radius = higher F1Smaller radius=smaller F1

Interpretations d0 = Vfront & Vback

d0 = Vfront = F2

d0 = Vback = F1

Stephen M. Tasko37

Interpretations r0 = F1 r0 = F1

When d0 (anterior) r0 = Vfront = F2 lip rounding = A/l = F1 & F2Stephen M. Tasko38

Another way to look at the dataStephen M. Tasko(Minifie, 1974)39

r0d0-+-+Articulatory Acoustic ComparisonsStephen M. Tasko

Traditional F1-F2 PlotF1-F2 Plot adjusted to reflectarticulatory space40Learning ObjectivesProvide an explanation for why we treat womens, mens and childrens vowels as equivalent even though absolute values of formants differ a lot.

Stephen M. Tasko

Stephen M. Tasko42

normalizing formant valuesStephen M. Tasko43

Clinical ExampleStephen M. Tasko44

Acoustic variables related to the perception of vowel qualityF1 and F2Other formants (i.e. F3)Fundamental frequency (F0)DurationSpectral dynamicsi.e. formant change over timeStephen M. Tasko45How helpful is F1 & F2?Data SourceHuman ListenersPattern ClassifierPeterson & Barney (1952)94.4 %74.9 %Hillenbrand et al. (1995)95.2 %68.2 %From Hillenbrand & Gayvert (1993)Stephen M. Tasko46How does adding more variables improve pattern classifier success?F1, F2 + F380-85 %F1, F2 + F080-85 %F1, F2 + F3 + F089-90 %Stephen M. Tasko47How about Duration?Nearby vowels have different durations

Stephen M. Tasko48

Stephen M. Tasko49What about Duration?

Stephen M. Tasko50What about Duration?

Some examplesStephen M. Tasko51What about formant variation?

Stephen M. Tasko52

What about formant variation?Stephen M. Tasko53

Naturally spoken /hAd/

Synthesized, preserving original formant contours

Synthesized with flattened formantsWhat about formant variation?Stephen M. Tasko54

Conclusion: Spectral change patterns do matter. What about formant variation?Stephen M. Tasko55

What do we conclude?Stephen M. Tasko56

Sinewave Speech Demonstration

Sinewave speech examples (from HINT sentence intelligibility test):Stephen M. Tasko57Selected issues that are not resolvedWhat do listeners use?Specific formants vs. spectrum envelopeWhat is the planning space used by speakers?ArticulatoryAcousticAuditory

Stephen M. Tasko58The important role of movementArticulatory movement = spectral change

Spectral change occurs as speakers transition within and between sound sequencesSpectral change plays a significant role inPerception of certain speech soundsOverall speech intelligibility

Stephen M. Tasko59DiphthongsSlow gliding (~ 350 msec) between two vowel qualities

ComponentsOnglide- starting point of articulationOffglide- end point of articulationArticulatory Transition = formant transition

Diphthongization: articulatory movement within the vowelVaries by geographic regionStephen M. Tasko60American English Diphthongs// - bye// - bough// - boy// - bay// - bow

Stephen M. Tasko61RESULTS

Original Duration:

96.0%

Neutral Duration:

94.1%

Short Duration:

91.4%

Long Duration:

90.9%

What can we conclude from all

this about how listeners recognize

which vowel was spoken?

1. Primary Cues:

F1 and F2 Relationships among the formants matter, not absolute formant frequencies

2. Cues that are of secondary importance, but definitely play a role in vowel perception:

f0 F3 (especially for //)

Spectral Change Patterns

Vowel Duration