Presentation for Processing Community Day Workshop in Los Angeles 2019 for the Accessibility, Disability, and Care track
Note: presentation included audio not included here.
3. Session Goals
1. Examine POPULAR MEDIA NARRATIVES around smart home
devices and feelings of CREEPINESS about synthesized
speech.
2. Learn about ACCESSIBILITY USE CASES for synthesized
voice output
3. Consider how IDENTITY is encoded in SOUND PATTERNS of
speech. Think about how to parameterize those patterns
4. Learn what’s possible with CURRENT TECH like the Web
Speech API & p5.speech. Explore future use cases for SONIC
EXPERIMENTATION - aka computers aren’t limited to the
constraints of the human vocal tract!
4. 1. What popular narratives
exist about speech tech? (e.g.
Alexa/Siri/Google Assistant as
creepy, sexist AI)
5.
6.
7. Not OK!!!
BUT. Let’s wait before
we throw out every
voice interface baby
with Alexa’s creepy
bathwater…
8. 2. What are other use cases
for synthesized speech
output? How is it used by
people with disabilities?
9. How a new technology is changing the lives of people who cannot speak –Jordan Kisner in The Guardian, 2019
10. Sara Young, recipient of “bespoke” digital voice from VoiceID
Image from https://twitter.com/LindsayMoran/status/679302950712422400
How a new technology is changing the lives of people who cannot speak
–Jordan Kisner in The Guardian, 2019
13. –Roger Ebert, Remaking my voice (TED2011)
“You all know the test for artificial
intelligence – the Turing test. A
human judge has a conversation with a
human and a computer. If the judge
can’t tell the machine apart from the
human, the machine has passed the
test. I now propose a test for
computer voices – the Ebert test. If a
computer voice can successfully tell a
joke and do the timing and delivery as
well as Henny Youngman, then that’s
the voice I want.”
14. Speech output as a
tool for non-visual
access
Screen reader demo with Sina Bahram!
https://www.youtube.com/watch?v=92pM6hJG6Wo
15.
16. “While synthetic speech may be difficult to
comprehend at first, its tolerance and
comprehension increases with more exposure
and experience.
…while most people would prefer a natural
voice, they find synthetic voices
acceptable for a number of applications…
When asked if they would like to choose
their own synthetic voice [95%] said “yes”…
[and some] suggested one of the voices they
use most often.”
– Text-to-speech audio description: towards wider availability of AD
Agnieszka Szarkowska, University of Warsaw (2011)
17. 3. What information is
encoded in speech? What
kinds of sound patterns
encode that information?
18. Listening Time!
Jot down anything you notice about the
speech sounds of these voices. Pay
attention to how they interact and change.
• Hey, youngblood. Lemme give you a tip. Use your White voice.
• Man, I ain’t got no White voice.
• Oh come on. You know what I mean. You have a White voice in there. You can
use it! It’s like when being pulled over by the police.
• Oh no, I just use my regular voice when that happens. I just say:
• “Back the f*ck up off the car and don’t nobody get hurt!”
• Aight. I’m just tryna give you some game. You wanna make money here?
Then read your script with a White voice.
From “Sorry to Bother You” (2018)
19. Let’s hear that
clip again
This time read by the closest available synthesized
speech voices. What do you notice now? How are the
clips different?
20. Computers are Social
Actors (CASA Theory)
Synthesized speech “mindlessly”
triggers social scripts like…
• Gender stereotyping: e.g. Female-
voiced computers rated more
informative about love, Male-voiced
computers rated more proficient in
technical subjects
• Reciprocity: When a computer voice
“helps” you, you’ll do more work for it
• Personality personification: Take
language cues to interpret personality.
Users like computers that match their
personality (similarity attraction)
https://en.wikipedia.org/wiki/Computers_are_social_actors
Clifford Nass, Jonathan Steuer, and Ellen R. Tauber. Computers are social actors. (1994, Stanford)
23. The Web Speech voices
available today
Umm…only one way to speak
English as a native US
speaker?
Even when systems have more,
all/almost all are in some flavor
of “Standard American” accent.
Google Chrome Web Speech API demo
24. What is the
“Standard American” (aka “General
American English”) Accent?
• White. Very white.
• From the Northeastern United
States in the very early twentieth
century
• Associated with wealthy highly-
educated White Anglo-Saxon
Protestant suburban
communities
• Characterized by the absence of
"marked" pronunciation features
of regional origin, ethnicity, or
low socioeconomic status.
https://en.wikipedia.org/wiki/General_American
Stock image of a newscaster
25. What are the effects of a
limited speech palette?
• What do these voices imply about “correct” and
“incorrect” ways of speaking?
• How does the use of a Standard American English voice
frame your relationship to it, and the tone of your
response?
• Is the voice a peer, superior, or subservient?
• Think about how you speak differently to family vs
friends vs strangers. Is a “conversation” with a computer
incapable of adapting its speech style even a
conversation?
27. “Segmental”
properties of speech
Fancy word for patterns of
pronunciation of vowels &
consonants. Examples:
• The pronunciation of the most
atomic units of sound in a
language
• e.g. “cot” vs “caught,” “pen” vs
“pin”
• “Butter” vs “Tammy”
28. “Suprasegmental”
parameters (aka prosodic)
Fancy word for patterns that extend
over syllables, words, and phrases.
Part of the grammar of a language.
Examples:
• Syllable structure
• Intonation (melodic pattern)
mapped to different syntaxes, e.g.
questions
• Stress (increase of volume and
duration within a word)
• Tone (variation in pitch)
29. “Paralinguistic”
properties of speech
Fancy word for miscellaneous speech stuff
that’s harder to codify and varies by speaker.
Examples:
• Position of sound in a space
• “Organic” vocal qualities - ie physiological,
size/proportion of speech organs
• Expressive vocal qualities - emotional tone
expressed through variations in loudness,
rate, pitch, pitch contour
• Filler words like “um,” “uh,” “like”
• “Backchanneling” - e.g. “mhm” in response
to another speaker
• Respirations: Breath, sighing, throat-
clearing
https://en.wikipedia.org/wiki/Paralanguage#Aspects_of_the_speech_signal
30. Speech Synthesis
Markup Language (SSML)
• Markup for prosodic features like
pitch, contour, pitch range, rate,
duration, volume
• Often embedded within VoiceXML,
the format used by telephony
systems
• Can add custom pronunciations
• Used by vendors like Amazon for
Alexa, Cortana
• Supposedly compatible with the
Web Speech API (though not
explored much, yet!)
31. Web Speech API
• https://github.com/mdn/web-
speech-api/
• Speech recognition and
speech synthesis
• Basic parameters for changing
rate, pitch, and voice
• Browser support on Chrome
• Available in Processing
through the speech.p5!
32. Coding Rainbow demo: https://www.youtube.com/watch?v=v0CHV33wDsI&t=34
p5.js-speech repo: http://ability.nyu.edu/p5.js-speech/
p5.speech
created for p5.js.
written by R. Luke DuBois
The ABILITY lab
New York University
p5.speech is a JavaScript
library that provides simple,
clear access to the Web
Speech and Speech
Recognition APIs, allowing
for the easy creation of
sketches that can talk and
listen.
33. 4. When speech is software, does
it even need to sound human?
What are other interesting sonic
possibilities?
34.
35. The Art of Noises
by Luigi Russolo (1913)
Six Families of Noises for the
Futurist Orchestra
1. Roars, Thunderings, Explosions,
Hissing roars, Bangs, Booms
2.Whistling, Hissing, Puffing
3.Whispers, Murmurs, Mumbling,
Muttering, Gurgling
4.Screeching, Creaking, Rustling,
Buzzing, Crackling, Scraping
5.Noises obtained by beating on
metals, woods, skins, stones, pottery,
etc.
6.Voices of animals and people,
Shouts, Screams, Shrieks, Wails,
Hoots, Howls, Death rattles, Sobs
Instruments for futuristic music, called "Bruitism",
partly electrically operated, built by Russolo, 1913
36. – The Art of Noises by Luigi Russolo (1913)
“[With machines] the variety of
noises is infinite…not merely in a
simply imitative way, but to
combine them according to our
imagination.”
37. Some more ideas for
WeIrD SPeEcH SyNtHeSiS!TM*
• A bunch of speech synthesizers walk into a bar / speech synthesizer cocktail party
• Drunk speech synthesizer
• Speech synthesizers loudly chewing, clearing their “throats” in an elevator
• Baby talk speech synthesizer
• Speech synthesizer with a speech impediment
• Code-switching speech synthesizer that changes pronunciations throughout speech
• SloMo Neil DeGrasse Tyson speech synthesizer
• Shy, socially anxious speech synthesizer
• Speech synthesizer which makes psycholinguistically realistic speech errors (e.g. spoonerisms
like “queer old dean” / “dear old queen”)
• Things that are physiologically impossible to articulate for a human!
* Not actually a trademark. (YET.)
38. NOW
Let’s design
some voices!
SUS examples http://research.nii.ac.jp/src/en/NITECH-EN.html
SUS generator https://github.com/ecooper7/SUSgen/blob/master/susgen.py
39. Thank you! Thanks to the Processing Community!
Contact: Emily Saltz, @saltzshaker / essaltz@gmail.com
44. Natural (humanoid) vs. Intelligible*
not at all humanoid indistinguishable from a human
not at all intelligible
fully intelligible
*DON’T FORGET TO ASK YOURSELF: “INTELLIGIBLE” TO WHO?
45. – Mariana Lin, Absurdist Dialogues with Siri (Paris Review)
“My fear in AI design is not the
singularity of global domination
(media, stand down … I think we’re far
from that); it’s the singularity of
conversational domination. I don’t
want AI to reduce speech to function,
to drive turn-by-turn dialogue
doggedly toward a specific destination
in the geography of our minds.”