Overview of CSSML

1
Overview of CSSML

• Yan Jun, Department Manager
• Anhui USTC iFLYTEK Co., Ltd
• University of Science Tech of China

2
Presentation Outline

• Motivation and solutions
• Standardization
• Application

3
CSSML

• Chinese Speech Synthesis Markup Language
• CSSML is a extension of SSML for Chinese
• Objective
• To meet Chinese speech synthesis requirements
• To provide more flexible and convenient methods
  to adjust parameters and optimize speech
  synthesis effect

4
Motivation

• Special problems of Chinese speech synthesis
• Pronunciation of Chinese characters
• Disposure of words composed of English letters
• Segmentation of Chinese words
• Requirements of Chinese speech market
• Using background music

5
Pronunciation of Chinese characters

• Syllables Chinese characters
• Chinese characters have four tones, or no tone to
  express unstressed syllables
• Chinese Romanization (PinYin) is widely used in
  China as a formal notation of Chinese character
  pronunciation.

?
guang
guang3
?
guang
guang1
6
words composed of English letters

• Words composed of English letters
• English words James, New York
• PinYin words Anhui, Hefei, Jiang Zemin
• PinYin words speak as English words
• Not according to pronunciation custom
• Difficult to understand

7
phoneme

• Attributes supported by the phoneme element are
  extended
• alphabet attribute can take py and ph attribute
  can be PinYin notation
• new lang attribute is added to indicate the
  language or dialect of the content

???
????Jiang Zemin
8
Segmentation of Chinese word

• Basic grammatical unit of Chinese Chinese
  character
• No blanks or punctuations to separate word
• Thus, one sentence may have several results of
  segmenting words that may be correct

???????
???????? The Bridge of the Yangtse River in
Nanking city
???????? Jiang Daqiao, the mayor of Nanking city
9
Segmentation of Chinese word

• Different result of segmenting words
• Greatly affect the meaning of the sentence
• The pronunciation of Chinese characters may be
  different ( monograph )
• Thus, influence or even destroy the effect of
  speech synthesis

???????? nan2 jing1 shi4 chang2 jiang1
da4 qiao2
???????? nan2 jing1 shi4 zhang3
jiang1 da4 qiao2
10
word and phrase

• word element is used to define the boundary
  between Chinese words
• phrase element define the boundary between
  phrases at different levels

???????
???????
? ????????
11
Using background music

• Synthesized speech can be played together with
  background music
• To upgrade user experience
• Background music may be added in a given position
• Background sound may be switched during the
  synthesis process

12
environment

• environment element is introduced to present the
  sound field environment of synthesizing
• src attribute
• repeat attribute

1.wav ???????????,?????????,??????????????????,
???????????????????????,?????????

13
CSSMLenterprise standard

• iFLYTEK setup the enterprise standard CSSML to
  define the markup language used in speech
  synthesis product in 2002
• Since 2003, the standard has been supported by
  InterPhonic product series of iFLYTEK

14
CSSML candidate of national standard

• Human-machine speech alternation standard
  workgroup of the Ministry of China Information
  Industry
• CSSML was proposed in the workgroup in 2003 and
  was widely debated
• CSSML was voted through by the workgroup in Oct
  24, 2005 and it will be submitted to the Ministry
  of China Information Industry as a candidate of
  national standard

15
Application

• Speech synthesis product that support CSSML are
  widely used in telecom, banking, insurance,
  negotiable securities, education and so on.
• telecom 168 and 114 information inquiry service
• securities stock comment, company introduction
• enterprise customer telephone service
• education to teach pronunciation of Chinese
  characters and words

16
Question?

• Thank you and good bye!