Topic 10: Text Analysis as part of TTS system

Question 1

Text to speech synthesis block diagram

Answer 1

text -> text analysis -> phonetic analysis -> prosodic analysis -> speech synthesis

Question 2

text analysis

Answer 2

includes preprocessing and conversion

document structure detection
text normalization
linguistic analysis

Question 3

phonetic analysis

Answer 3

grapheme-to-phoneme conversion

Question 4

prosodic analysis

Answer 4

pitch and duration attachment

Question 5

speech synthesis

Answer 5

voice rendering

Question 6

Document structure detection detail

Answer 6

why needed? there are various text formats
TTS don’t pay attention to structure, bottom line is to synthesize speech

more tagging better speech expression

SSML

Question 7

Text normalization detail

Answer 7

different NLP have different normalization purpose

For TTS, needed to be done until text converted to readable form

used to overcome ambiguity to certain extend

Why normalize?

• symbols
• number format
• combination of both
• abbreviation and acronym
• emoji

Question 8

Normalisation: Abbreviation and

Acronym Expansion

Answer 8

example steps

Question 9

Normalisation: Pattern Matching in RE
when to use RE?
where it can be done
what is RE?
what is string

Answer 9

• When to use RE: search and modify
• Where it can be use in: string, pattern, corpus matching
• A regular expression, often called a pattern, is an expression used to
specify a set of strings required for a particular purpose.
• String: For text-based search, a string is any sequence of
alphanumeric characters

Question 10

More example application of RE

Answer 10

• test for pattern
• replace text
• extract substring

Question 11

Linguistic Analysis (LA) detail

Answer 11

Also known as syntactic and semantic parsing in NLP

Information desired for TTS from parsing analysis:
o Word part-of-speech (POS) or word type
o Word sense
o Phrasal cohesion of words: idiom, syntactic phrases, clauses, sentences
o Modification relations among words
o Anaphora (co-reference) and synonymy
o Syntactic type identification: questions, quotes, etc.
o Semantic focus identification (emphasis)
o Semantic type and speech act identification: requesting, informing, narrating,
etc.
o Genre and style analysis

In principal we don’t need all for these for a TTS, but we need those that can provide TTS-specific functionality

LA supports the phonetic analysis and prosodic generation phases
what is needed for TTS
- sentence breaking/tokenizer
- POS tagging
- homograph disambiguation
- noun phrase and clause detection
- sentence type identification