Designing Text to Speech with TTS Discover X

1
Designing Text to Speech with TTS Discover X

• Daniel Woo
• University of New South Wales

2
Introduction

• Speech integration will help make your
  application a true first class citizen of Mac OS
  X
• Text-to-Speech (TTS) can be fine tuned using
  punctuation, embedded speech commands and
  phonemes
• Introduce TTS Discover X

3
Accessibility

• Speech is key component of Universal Access
• Support for visually impaired users, users with
  reading difficulties
• Supporting speech will help make your application
  more accessible
• Any visible text from Aqua controls can be spoken
  with text to speech

4
User Initiated Speech Synthesis

• When can the user initiate spoken text?

5
Edit Menu
6
Context Menu
7
Services Menu
8
Press a key Text under mouse
9
Computer Initiated Speech

• Talking Alerts
• Your application

10
Talking Alerts

• Message title text of an alert box can be spoken
• Ideal when attention is not focussed on the
  screen
• User defineable delay time
• Different voices

11
Developer Initiated Speech

• AppleScript
• Say Hello, George saving to file
• Speech Manager
• CopyCStringToPascal (theStr cString, dst)
• SpeakString(dst)
• Improved Cocoa Support
• Panther

12
Application Developers

• Consider how your application sounds when spoken
• When will your application speak?
• How do your custom controls work with speech?
• Universal Access support

13
Controlling Speech Synthesis

• Words - English
• Punctuation
• Emphasis
• Phonemes
• Pitch
• Rate

14
! , . ?

• Punctuation in text material will produce
  appropriate pauses (.,) and pitch rises (?!)
  and falls (.)
• Pitch variation (intonation) conveys meaning

15
Punctuation Demonstration

• How are, you!
• How are, you?

16
Embedded Speech Commands

• Used when punctuation is not enough
• The TTS engine detects embedded mark up tags in a
  text string
• nmbr LTRL
• emph emph-
• inpt PHON

17
Numbers

• call me on extension 2003
• call me on extension nmbr LTRL 2003

18
Emphasis

• Focus the users attention on important information

19
Emphasis Demo

• Would you like me to wash the car?

20
emph

• Would you like emph me to wash the car?
• Would you like me to emph wash the car?
• Would you like me to wash the emph car?

21
Phoneme Definitions

• inpt phon w1UHd yUW l1AYk mIY tUW w1AOS
  DAX k1AAr
• The correct way to create words that are not in
  the dictionary
• Specify the pronunciation in terms of phoneme,
  stress, emphasis etc.

22
TTS Discover X

• Original design goal was to help reduce the need
  to remember phoneme symbols
• Provide a tool to quickly test and modify
  pronunciation of TTS strings

23
Features

• Enter sentence text
• Listen to spoken text
• Convert to phoneme representation
• Modify phoneme representation
• Listen to phoneme text to speech
• Display a phoneme keyboard

24
Support for Non-Speech Users

• Tooltips to reveal example words
• Audio preview the phoneme, with an example
  (optionclick)

25
Demonstration
26
Phonetic Fonts

• Phonetic font support is available out of the box
  in Mac OS X via Unicode
• Previously, specific International Phonetics
  Association (IPA) fonts needed to be installed
• eg. SILManuscript, SILSophia fonts

27
Forthcoming Features

• Additional speech mark up tags
• IPA keyboard
• IPA support

28
Repeat After Me

• Extracts the pitch contour from real speech and
  applies it to a corresponding text to speech
  utterance
• Help developers fine tune pitch

29
Resynthesis

• Standard TTS
• Human recording
• Resynthesized TTS

30
New Tune Speech Tag

• inpt TUNE
• Specify parameters
• Phoneme
• Duration
• Time varying pitch
• Fine tune utterances where appropriate

31
Voices

• Based on US English
• Formant Synthesis
• Macintalk 3 Voices
• Entertaining voices
• Bubbles, cello
• High Quality Voices
• Victoria, Bruce, Agnes
• New Victoria in Panther
• Third party voices

32
Usability Testing
33
Conclusions

• Speech synthesis can be used out of the box with
  high degree of functionality
• Fine tune the synthesized utterance to convey the
  intended meaning

34
Conclusions

• A greater degree of control is available using
  embedded speech commands and phoneme symbols
• Improved control over pitch and duration will be
  made available

35
Conclusions

• Adopt accessibility
• Test and Evaluate

36
Conclusions

• TTS Discover X was designed to help non-speech
  people quickly start discovering speech synthesis
  on Mac OS X
• Other educational applications in linguistics and
  phonetics

37
QA