Adventures in Tech - Why Apple can't get Siri right
Adventures in Tech - Why Apple can't get Siri right
2014-04-22
if you were a geeky kid you probably
dreamed of a computer that would obey
your voice commands Apple Siri brings
that dream to life so why isn't it very
good
in 2011 Apple revealed the
highest-profile voice assistant yet but
anyone who's used Siri will know it
often gets it wrong misunderstands or
fails to perform as advertised what is
it that makes series job so complicated
and what does Siri needs to improve
series first job is speech recognition
also known as ASR it breaks down audio
into bite-sized sounds which get checked
against a vast database of spoken
language the process is extremely
sophisticated though because it
processes data online the system pulls
down if you've got no reception and as
Patrick Naylor from London Imperial
College explains it can be thrown off
they're going to find it very hard to
recognize anything which isn't already
contained in the training data so if
they come across unexpected accents or
unexpected noise sources in the audio
then it doesn't recognize them audio
that doesn't match the database won't be
recognized
that's why Siri doesn't always handle
accents well and can make major mistakes
in noisy chaotic environments believe it
or not speech recognition is the
relatively easy bit the second half of
what Siri has to do is a tough a problem
not just figuring out what you say but
what you mean
so this is natural language
understanding and LU where we're really
trying to do something useful with the
information that we've obtained Siri has
to deduce the intent behind your words
if you ask Siri for two tickets to a top
grossing film at 9:00 p.m. within a mile
of your house it has to identify from
that jumble of words that what you want
is to book movie tickets as humans way
greater grasping the subtext beneath
simple strings of words it's how we
understand each other
getting Siri to listen like a human is
about much more than simple speech
recognition so the natural language
understanding parlor
is a much harder problem and it's with
that than most people I think feel
dissatisfied at the current time from
these services which are available Nou
is a relatively new issue compared with
speech recognition and the mainly ways
to make Siri a bit more human in future
for instance Siri could be more
conversational if you ask for directions
to a New York address Siri could ask you
if you know the way to Brooklyn Bridge
and offer to direct you from there
talking back could make Siri more
powerful if Siri was able to develop his
dialogue capability beyond what it has
at the moment which is that which is not
non-existent but relatively primitive
then that would make us feel I think
much more able to get the information
that we want in the way that we want it
next theory relies on a microphone but
what if it also used your phone's camera
to monitor your lips and gather extra
speech data this could filter out
background noise or maybe even help Siri
understand us when we've had one too
many white wine spritzers
say I need you know everything you could
tell me about monkeys I seriously cannot
wait for the robot uprising finally Siri
performs better the more it knows if we
want futuristic voice controlled
assistance the price we pay may be our
privacy Siri is more useful once it
knows your address imagine what it could
do with access to your medical records
or an intimate knowledge of your
finances we may reach a point where the
only way to make systems like Siri more
useful is to let them into our private
lives whether we're comfortable with
that is something we'll need to decide
and soon so we've got to get a match
between what are our expectations and
what are our concerns particularly about
privacy and how do we align those
because we can't have it both ways do
you think Siri will get better or not
let me know and check back next time for
another adventure in tech
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.