|
SAN JOSE, Calif.
— YouTube said Thursday that it will start
automatically captioning videos on its site, opening up
a huge share of its content to people who are
hearing-impaired, and a first step in creating a network
of videos that could be subtitled between many
languages.
For now,
YouTube can only transcribe videos uploaded in English
into text, which can then be translated into text
captions in other languages. But given the 20 hours of
video content that YouTube uploads each minute, the
San Bruno
unit of
Google
said this is the largest effort ever tried on the
Internet to harness evolving speech recognition
technology to caption video content.
Students
and teachers at the
California Academy for the Deaf
in
Fremont
who attended Thursday's announcement at YouTube said the
new service could be revolutionary, in terms of helping
deaf people feel connected to a society that
increasingly communicates and expresses itself through
video.
Watching
video on YouTube or elsewhere, "we felt like we
weren't a part of the world," said 17-year-old
Angel Harrington
of the Academy, speaking to reporters through sign
language. "But now we have communication
accessibility. We really can understand what is going
on, and we do feel like we're on an equal playing
field" with hearing people.
While
YouTube has offered closed captioned video in the past,
the captions generally had to be provided by the video
producer, and were a very small share of YouTube's
database. With the new service, which launched Thursday,
any video that meets sound and technical quality
standards automatically will be captioned by
Google's
speech recognition technology.
For
Ken Harrenstien
, a
Google
software engineer whose business card bears the
unofficial title "Caption Jedi," the new
service is the culmination of five years of intense
effort. Harrenstien, who is deaf, said he never attended
classes at
MIT
, because "there was no point in me sitting there
and watching peoples' mouths move."
But the
new caption technology has allowed him to go back and
view captioned videos on YouTube of those
MIT
lectures he missed years ago.
"This
is huge," Harrenstien said. "It's what I've
dreamed of for so many years. You can go to (almost) any
video online, and expect to see some captions."
Speech
recognition has been a focus of
Google's
for the past five years, and the YouTube technology is
similar to what
Google
uses to transcribe audio voice mails through its Google
Voice service, and to provide spoken Web searches from
smart phones.
But it is
not yet a mature technology. Engineers said Thursday
that the machine translations are sure to contain some
funny mistakes, showing a YouTube video where a
Google
engineer says "included in every device will be a
SIM card" as he shows off a new smart phone, but
which the speech transcription rendered as
"including in every device will be a salmon."
"We
know it's not perfect, and sometimes it will be
funny," Harrenstien said. "But it's better
than nothing."
Mike Cohen
, manager of
Google's
speech technology group, said speech recognition has
evolved over the past 50 years, but that a convergence
of technology has only recently made it good enough for
automatic transcription at the scale of a YouTube.
While
computers still struggle to transcribe accents and cope
with background noise, accuracy will continue to improve
over time, Cohen said. He said captioning of languages
other than English is a "very high priority"
for
Google
.
Administrators
at
Stanford
and the
University of California-Berkeley
said the service could make lectures and other
instructional content that they post on YouTube
available to audiences around the world.
"Blowing
through those doors and language barriers that exist,
making it easier for folks to understand when they are
non-native English speakers, it's really a huge benefit
for that," said
Ben Hubbard
, manager of webcast.berkeley.edu, which provides free
video content of
UC-Berkeley
courses.
|
