12-Hirschberg_Chew

Report
Prosody Modeling (in Speech)
by Julia Hirschberg
Presented by Elaine Chew
QMUL: ELE021/ELED021/ELEM021
26 March 2012
Sources
• Hirschberg, J. (2012). Lecture notes on Prosody
Modeling (draws from Hirschberg, J. (2011).
Interspeech Tutorial - More Than Words Can Say)
– http://www.cs.columbia.edu/~julia/courses/CS4706/r
epresenting.pptx
• Hirschberg, J. (2004). Pragmatics and Intonation.
In L. R. Horn and G. L. Ward (eds.): The handbook
of pragmatics, Blackwell Publishing Ltd., 515-537.
• Hirschberg, J. and M. E. Beckman (1994). ToBI
Labeling Conventions.
• Cardinal examples:
– http://www.cs.columbia.edu/~agus/tobi
Prosodic Variation and Interpretation
• Prominence: accents, stress
– John only introduced MARY to Sue
– John only introduced Mary to SUE
• Boundaries: disjuncture between words
– Bill doesn’t drink | because he’s unhappy
– Bill doesn’t drink because he’s unhappy
Example 1: Prominence
Example 2: Phrasing
ToBI Goal
• Capture enough variation to explain
similarities and differences in prosodic
meaning
ToBI Scheme
• ToBI annotation tiers:
– Orthographic tier: Time-aligned words
– Break-index tier: degrees of junction (0=no
boundary; 4=full intonational phrase boundary)
– Tonal tier: pitch accents, phrase accents, boundary
tones
– Miscellaneous tier: disfluencies, non-speech
sounds, etc.
ToBI Break Indices
•
•
•
•
Level 0: no boundary
Level 1: word boundary
Level 2: Strong juncture with no tonal boundaries
Level 3:
– minor or intermediate phrase
– Consists of >=1 pitch accent(s), aligned with stressed syllable or
lexical items (phrase accent)
– Phr accents describe movement to phrase boundary: H-, !H-, L-
• Level 4:
– major or intonational phrase (associated with tonal tier
describing phrase accents and boundary tones for each level)
– Consists of >= 1 Level 3 phrase(s) plus high/low boundary tone
(H% or L%) at the right edge of phrase
– Boundary tones describe pitch movement immediately before
boundary
Standard Declarative Contour
Ends with L- L%
Example: H* L- L%
Standard Yes-No Question Contour
Ends with H- H%
Example: L* H- H%
Phrase Ending Types
L-L%
L-H%
H-H%
H-L%
!H-L%
Break Indices Differences
• Associated with
– Variation in f0
– Phrase final lengthening
– Glottalization
– Some amount of pause
Pitch accents: Intonational Prominence
• Achieved through
– Different tone targets
– Differences in f0 height
– Being louder and longer
• Hierarchy
– Last accented word tends to be most prominent
– Most prominent accent in intermediate phrase is
called phrase’s nuclear accent or nuclear stress
ToBI Accents
H*
• H*: simple high (declarative)
L*
• L*: simple low (yes-no question)
L*+H
L+H*
H+!H*
• L* + H: scooped, late rise
(uncertainty, incredulity)
• L + H*: early rise to stress
(contrastive focus)
• H + !H*: fall onto stress (implied
familiarity)
(* indicates stressed syllable)
H*
• Most common accent in American English
• Simple peak in f0 contour
• Typically found in standard declarative
utterances
• Commonly used to convey accented item
should be treated as NEW information
L*
• Modeled as valleys in f0
• Conveys accented item is salient but not part
of what is being asserted
• Typically characterize prominent items in yesno question contours
• Often employed to make initial prepositions
or adverbs prominent, or to mark discourse
readings of cue phrases
L + H*
• Can be used to produce a pronounced
contrastive effect
• Example:
The Smiths aren’t inviting anyone important
– They invited L + H* Loraine
(contradicts initial claim that Loraine is unimportant)
– They invited L* + H Loraine
(uncertainty about whether Loraine is an important
person)
L + H*
H + !H*
• Fall onto stressed syllable
• Associated with implied sense of familiarity
with the mentioned item
• Example (“reminding” case):
– A: No German has ever won the Luce prize
– B: H + !H* Joachim’s from Germany
H + !H*
More on !H
• Downstepped accent
– !H*: Half the job is accomplished by just starting it
– L + !H*: There’s a lovely one in Bloomingdale’s
– L* + !H: Don’t hit it to Joey.
Example (Praat)
L-L%
L-H%
H-L%
H-H%
H*
L*
L*+H
Examples from http://www.cs.columbia.edu/~agus/tobi
L-L%
L-H%
H-L%
H-H%
L+H*
H+!H*
H* !H*
Examples from http://www.cs.columbia.edu/~agus/tobi
ToBI Family
•
•
•
•
•
•
•
•
American English
German
Japanese
Korean
Mandarin
Portuguese
Greek
Catalan
Exercise (3)
• Anna frightened the woman | with the gun
• Anna frightened | the woman with the gun
• Who held the gun in each case?
Exercise (4)
• Mary knows many languages you know
• Mary knows many languages | you know
Exercise (7)
• John laughed | at the party
• John laughed at | the party
Exercise (11) (12)
• We only suspected | they all knew that a
burglary had been committed
• We only SUSPECTED | THEY all KNEW | that a
BURGLARY had been committed

similar documents