Computational phonology and Texttospeech

1
Computational phonology and Text-to-speech

• Presenter Jessie Ester

2
Introduction

• Computational phonology
• ?Automatic Speech Recognition (ASR)
• take an acoustic waveform as input and produce
  as output a string of words.
• ?Text-To-Speech (TTS)
• take a sequence of text words and produce as
  output an acoustic waveform.
• ? How words are pronounced in terms
• of individual speech units called
  phones.

3
4.1 Speech sounds and phonetic transcription

• A phone a speech sound, represented by IPA or
  ARPAbet.
• IPA An evolving standard with the goal of
  transcribing the sounds of all human languages.
• ARPAbet A phonetic alphabet designed for
  American English using only ASCII symbols.

4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
4.2 The phoneme and phonological rules

• A phoneme /t/

8
4.3 phonological rules and transducers

• Transducer of phonological rule

9

• Many different phonological rules apply between
  the lexical form and the surface form.
• Underlying form surface form
• Sometimes these rules interact the output from
  one rule affects the input to another rule.
• One way to implement rule-interaction in a
  transducer system is to run transducers in a
  cascade.

Phonological rule
10

• iz
• English plural suffix /z/ s
• z
• Insertion rule and devoicing rule
• These two rules must be ordered
• Feeding
• Bleeding
  

11

• Two-level morphology (Koskenniemi,1983)
• Most phonological rules are independent of each
  other.
• It is more efficient to run phonological rules in
  parallel than in series.
• Two-level rules can be thought of as a way of
  expressing declarative constraints on the
  well-formedness of the lexical-surface mapping.

12

• Four rule operators

13

• In order to avoid feeding and bleeding, two-level
  rules must have the ability to represent
  constraints on two levels.
• The use of
• ab ?c__ a is realized b after a surface c
• ab ?c__ a is realized b after a lexical c

14

• Koskenniemis two-level rules finesse the issue
  of ordering by potentially referring to both
  underlying andsurface forms.

15
(No Transcript)
16
(No Transcript)
17
4.4 Advanced issue in computational phonology

• The Yawelmani dialect of yokuts
• ? Harmony a surface V changes its form to look
  
• like a neighboring V.
• Constraint only when the stem V and the surface
  V are of the same height.

18

• ? Lowering causes long high vowels to
• become low. /i/?/e/, /u/?/o/
• ? Shortening shorten long vowels if they
• occur in closed syllables.

19

• Order of the rules
• harmony? lowering? shortening

20
Optimality theory (OT)

• OT (Prince Smolensky) involves the idea of
  competing constraints, which can be ranked in
  importance with respect to each other. Due to
  this ranking, a less important constraint can
  sometimes be violated in order to obey a more
  important constraint.
• GEN takes an underlying form and produces all
  possible surface forms.
• EVAL consists of a set of ranked constraints
  (CON) and an algorithm for choosing the best
  candidate.

21

• Yawelmani three consonants in a row (CCC) are
  not allowed to occur in a surface form.
• ?CCC ? C-deletion
• C?e/C__C /hnil/
• ?CCC ? V-insertion
• e?V/C__CC /hin/

22

• Yawelmani dont allow complex onsets or complex
  coda.

23

• Can a derivation in OT be implemented by
  finite-state transducer?
• If
• GEN is a regular relation (assuming the input
  doesnt contain context-free trees of some sort)
• The number of allowed violations of any
  constraint has some finite bound.
• then an OT derivation can be computed by finite
  states means.

24

• Merciless cascade each constraint is implemented
  as a filter transducer which lets pass only
  strings which meet the constraint.
• Lenient cascade it is essential to only enforce
  a constraint if it does not reduce the candidate
  set to zero.