More on this book
Kindle Notes & Highlights
by
Cathy Pearl
Read between
July 3 - August 11, 2019
user-initiated interaction
timeout is for no speech detected (NSP).
prerecorded video (http://bit.ly/2hcpvv4/).
why the NSP timeout might have been triggered. First, although the system thinks it did not hear any speech, it could be mistaken.
too-much-speech (TMS).
recognizer is triggering on nonsalient speech, and you need to determine why.
recognition engine does not typically return just one result for what it thinks the user said. Instead, it returns what’s referred to as an N-best list, which is a list of what the user might have said, ordered by likelihood
(usually
How do we get access to this data?
If your VUI takes advantage of the N-Best list, however, you can put rejects on a skip list;
VUIs.
It blows me away that this is not more common. Aside from volume, it is usually not difficult to tell 2 different voices apart. It's not revolutionary technology
which device should respond when I make a request?
much less data for that type of speech
can
This is confusing to a lot of users who are used to ivr and similar systems
fail.
Not necessarily -- you could account for this
Speech recognition refers to the words that the recognition engine returns; NLU is how you interpret those responses.
Constrained Responses
Open Speech
(myself, Mark Anikst, and Lisa Falkson) developed at Volio
credibility.
I don't understand this.
the importance of looking out for negative responses,
(these bits are sometimes referred to as slots)
impression.
Give an example
conversation.
That's exactly why if you're mapping, you should display the mapped value
MPQA negative and positive lists,
Open source tools such as KNIME
Affectiva
Beyond Verbal
desire.
The same is true for people. If people can be trained, so can computers.
Speech Synthesis Markup Language (SSML),
pronunciation dictionary
Balogh’s paper[29] as well as Chapter 11 of Voice User Interfaces Design.
Speaker Verification
Charles Schwab recently rolled out voice ID to authenticate account logins.
speaker identification
ethical.
It's arguably not ethical to be listening all the time in the first place. The echo is forming new norms when it comes to this.
(which should of course be anonymized).
What does anonymous mean in this context?
week.
This could also be considered rude
long list
make their users more comfortable with doing transactions on their mobile devices,
DEBORAH DAHL, MOBILE VOICE 2016
Hound has a knack for responding to multiple queries in one.
O’Reilly Media has an abundance of user research video tutorials available online, which you can access at http://bit.ly/user-research-fundamentals-lp.
Jennifer Balogh,
Latin Square design
Jakob Nielsen
User Testing Blog at https://www.usertesting.com/blog/2015/01/29/screener-questions/.
UserTesting (https://usertesting.com/) and UserBob (https://userbob.com/),
Likert, L., and Algina, J.:[40]
Intelliphonics,
Robbie Pickard while at Volio.
