Model of Phonological Representation (Articulatory Phonology)

Transcription

1 Model of Phonological Representation (Articulatory Phonology) Phonological representation of an utterance is a gestural coupling graph: NODES specify gestures: context-invariant tasks for VT constriction devices LIPS TT (Tongue Tip) TB(Tongue Body) VEL (Velum) GLO (Glottis) EDGES specify coordination among gestures: relative phase targets for pairs of gestural oscillators Each gesture is associated with a timing oscillator Timing oscillators trigger the activation of their associated gesture(s)

2 Topology of coupling graph defines syllable structure Coupling model of the syllable Ons gestures in-phase to Nucleus gestures Coda gestures anti-phase to Nucleus gestures Oral Constriction gestures within Ons or Coda eccentric phases to each other Ons sp GLO(wide) TT (crit alv) σ Rime Nuc Coda E d sped TT (clo alv) LIPS (clo) TB (wide pal)

3 Example Coupling Graphs bad VEL(wide) mad LIPS (clo) TT (clo alv) LIPS (clo) TT (clo alv) TB (wide phar) TB (wide phar) GLO(wide) sad GLO(wide) sped TT (crit alv) TT (clo alv) TT (crit alv) LIPS (clo) TT (clo alv) TB (wide phar) TB (wide pal)

4 Stages of Speech Production Model Planning Gesture oscillators all start at random phases. Over repeated cycles, coupling forces cause oscillators to settle at stabilized relative phases (Saltzman & Byrd, 2000). Cycles of stabilized oscillations used to determine times of gestural activations and deactivations (gestural score) Constriction Formation Coordinated motion of articulators results from activations of invariantly specified gesture tasks.

5 TaDA input and Output TVten.o PHten.o ten ten.wav TVten.g Figure 1. Information flow through TADA models.

6 Example: /spœt/ Lips CLO REL TB PHAR WIDE TT ALV CRIT ALV CLO GLO WIDE 50 ms Time

7 TADA files Gestural score file (tv<id>.g) Coupling graph files tv<id>.o ph<id>.o Text Input English Orthography cathie#best Must be in lower case Real English words Generates American English pronunciation ARPABET (syllabified) (KAETHIY)#(BEHST) Can be used for non-words

8 Contents of TV files Gesture Specifications in TV.o TV_name Osc_ID targ f d art_wts alpha beta Gesture Specifications in TV.g TV_name 0 begfr endfr 0 targ f d art_wts alpha beta 8

9 TaDA variable names in Mview TVs Pellets Articulators LA UL (x and y) LX lip protrusion LP LL (x and y) UY upper lip y TTCD TT (x and y) LY lower lip y TTCL TF (x and y) JA jaw angle TBCD TD (x and y) CL tongue body length (roughly front/back) TBCL TR (x and y) CA tongue body angle (roughly high/low) VEL JAW TL tongue tip length GLO TA tongue tip angle F0 Preface TVs with G to get gestural activations 9

10 Automatic Computation of Coupling graph Orthographic Input Syllabified CMU Dictionary Syllabified Gestural ARPABET Dictionaries Coupling LINKS + Activation Phases (PH.o) Gestural regimes + Oscillator ID = NODES (TV.o) Syllable Coupling Model

11 Gestural Dictionaries: ARPABET to gestures ARPA Organ Osc TV Constr B Lips clo LA CLO B Lips rel LA REL B Velum clo VEL CLO P Lips clo LA CLO P Lips rel LA REL P Glottis h GLO WIDE P Velum clo VEL CLO M Lips clo LA CLO M Lips rel LA REL M Velum n VEL WIDE Etc.. ARPA Organ Osc TV Constr D TT clo TTCL ALV D TT clo TTCD CLO D TT rel TTCL REL D TT rel TTCD REL D Velum clo VEL CLO T TT clo TTCL ALV T TT clo TTCD CLO T TT rel TTCL REL T TT rel TTCD REL T Glottis h GLO WIDE T Velum clo VEL CLO N TT clo TTCL ALV N TT clo TTCD CLO N TT rel TTCL REL N TT rel TTCD REL N Velum n VEL WIDE

12 Gestural Dictionaries: Gestural Control Regimes TV Constr Target Alpha LX JA UH LH CL CA TL TA NA GW TTCL DENT TTCL ALV TTCL ALVPAL TTCL PAL TTCL REL TTCD CLO TTCD CRIT TTCD NAR TTCD REL

13 TADA outputs Model articulator time functions Constriction (Tract Variable) time functions Pseudo-sound Input file to HLsyn (<id>.hl)

14 Generation of Acoustic output Model Vocal Articulator Tract motions Model Tract Variable motions Area Functions Resonance Model Formant Frequencies Bandwidths HLsyn Acoustic Output

15 Editing Capabilites Gestural score editing in TADA Graphical Sliding gestures (or groups) in time Stretching or shrinking activation intervals Deletion of gestures Addition of gestures Numerical Dynamical parameters Blending parameters