LING124 Prosody in TTS

1
LING124 Prosody in TTS

• November 18, 2008

2
Prosody

• Use of suprasegmental features to convey
  sentence-level pragmatic meanings
• F0, duration, energy
• Difference between statements and questions
• Salience of particular words or phrases
• Affective and emotional meaning
• Phrasing, prominence, tune

3
Phrasing (1)

• Intonational phrase
• I wanted to go to London, but could only get
  tickets for France
• Intermediate phrase
• I wanted to go to London
• Final vowel is longer at the end of phrase
• Pause after the final word in the phrase
• Slight decrease in F0 from the beginning to end
  of the phrase

4
Phrasing (2)

• Hand-crafted rules
• e.g. Insert a phrase break after punctuation
• Features for machine learning classifiers
• Number of words and syllables in the sentence
• Distance from the beginning and the end of the
  sentence
• Distance from the previous phrase break
• Distance from the last punctuation
• Part-of-speech tags of neighboring words

5
Prominence (1)

• Pitch accent
• Im a little surprised to hear it characterized
  as upbeat
• Emphatic accent
• I know SOMETHING interesting is sure to happen
• Unaccented
• Reduced

6
Prominence (2)

• Content words, new or informative words are more
  likely to bear pitch accent
• New or informative words
• Low N-gram probability
• Term-frequency inverse-document frequency
• Term-frequency
• Relatively how often a word occurs in the given
  document
• Inverse-document frequency
• Log of inverse of the proportion of documents
  that contain the word

7
Tune

• Rise and fall of F0 over time
• Final-fall
• Question-rise
• Continuation-rise (when nouns are separated by
  commas)

8
ToBI (1)

• Prominence and tune
• Pitch accents and boundary tones
• Pitch accents
• H Peak accent
• L Low accent
• LH Scooped accent
• LH Rising peak accent
• H!H Step down

9
ToBI (2)

• Boundary tones
• L-L Final fall
• L-H Continuation rise
• H-H Question rise
• H-L Final-level plateau

10
ToBI (3)

• Prosodic structure
• Intonational phrase break 4
• Intermediate phrase break 3
• Pause between words 2
• Word boundaries without pause 1

11
Duration - Klatt (1979)

• Determine by how much the mean duration of a
  phone should be lengthened or shortened
• Duration (inherent duration minimum
  duration)A minimum duration
• Features (A), from Taylor (2009)

12
Duration Klatt (1979) (2)
13
F0

• Specify F0 target points for each pitch accent
  and boundary tone and interpolate among these
  targets
• Target points are specified in terms of pitch
  range
• Baseline frequency Lowest F0 in utterance
• Topline Highest F0 in utterance
• Specify where in the accented syllable the
  targets apply
• Example rules from Jilka et al. (1999)
• H 100, L 0, LH 20-100,H-H 120,
  L-L -20
• H on the 60 of the way through the voiced part
  of the accented syllable