Title: When Voice and Content Dont Match'''
1 - When Voice and Content Dont Match...
- Gabriel Adauto
- Ulla Foehr
- Megan Huddleston
- Michael Somoza
2Key Questions
- Is emotion attributed to synthetic voices?
- How does voice emotion affect
- perceptions of the content?
- perceived suitability of the content?
- How does the match/mismatch between voice
characteristics and content affect - liking of the information?
- credibility of the information?
3Importance of the Questions
- Psychological Issues
- Good/bad is a fundamental psychological category
- Media Equation If human voice transmits emotion,
so does synthetic voice - Can one simulate emotion?
- Consistency expectations is a fundamental
psychological principle
4Importance of the Questions
- Design Issues
- Usually one voice, regardless of content
- Detection of emotion in text content is unsolved
problem - How important is this?
5Context
- Telephone system
- Varied Topics
- News
- Movie Descriptions
- Health Information
- Happy and sad content in each category
- Web Questionnaire
6Procedure
- 1. Participants recruited by email and directed
to a website - 2. Participant goes to a website for instructions
- 3. Participant calls the experiment and enters a
group code to hear content - 4. After hearing each story, the participant
answers questions on the Web
7Manipulations
- TTS vs. Recorded (between)
- Match/mismatch of voice emotion and content
(between) - Two stories per topic (within)
8Demo of the Stimulus
TTS Voice Sad News Another dead gray whale
washed ashore in San Francisco Bay yesterday,
deepening a mystery that is baffling scientists
about the recent spate of deaths.
Recorded Voice Happy Health For the first time,
scientists say gene therapy has unequivocally
succeeded. Doctors used the treatment to save the
lives of several infants who have a severe immune
disorder.
9Interest in Topics
Extroverts more interested
Introverts more interested
10Perceived Content Valence
11Liking of the Content
12Credibility of the Content
13Psychological Implications
- Even synthetic speech can manifest emotion
- Extroverts will like it more
- Emotion of voice influences perception of content
- Even in TTS!
14Psychological Implications
- Consistency has important psychological
consequences - Great liking
- Less credibility
- Liking/credibility tradeoff
15Design Implications
- Decisions of whether to match or mismatch voice
and content is crucial - Emotion is important even for non-emotional
machines - Voices should be cast based on
- Content
- Target audience
- Products on site (party items vs. funeral home)
- All modalities should be consistent
- Text
- Voice
- Facial expression
16Open Questions
- What other voice characteristics must be
consistent with content? - Gender
- Age
- Arousal
- Do synthetic faces manifest emotion? What about
synthetic bodies?