Broadcast Beat Magazine 2018 NAB Show Edition | Page 94
captioning solutions for years.
ASR offers broadcasters an opportunity to train
the engine in the short term and reap the rewards
in the long term as accuracy increases with addi-
tional content.
Machines to the rescue
Machine learning is bringing a new dawn of pos-
sibilities to the captioning scene. A new genera-
tion of cloud-based AI and ML systems are able
to analyze, understand and learn the surrounding
context of dialog content in much the same way
that humans do. Such systems have a chance to
succeed where previous captioning-automation
platforms have fallen short. Harnessing the power
of the cloud, AI-based automated speech recog-
nition (ASR) systems are now able to examine
and interpret words. Using sophisticated machine
learning algorithms, cloud-based ASR engines
can help do the transcription work faster and can
learn to become more precise.
With the ability to self-learn from every correc-
tion, recognition accuracy of cloud-based ML
automated speech recognition systems improves
with each use. Names and proper nouns are
automatically extracted from human-reviewed
transcription and used as glossary words, ensur-
ing they will be recognized and spelled properly
in subsequent uses. The key element behind the
efficacy of ASR for video content is a machine
learning from a human to perform better. And
as the accuracy of auto-transcription increases,
humans can leverage advanced ASR systems to
increase their own productivity
When the unlimited horsepower of the cloud
is leveraged against the enormous amounts of
data being processed and learned from diverse
sources, on-the-ground solutions cannot hope
to compete. The next generation of cloud-based
ASR systems are already working with dozens of
languages, and they continue to evolve, learn and
adapt as needs change.
Missed Opportunity:
Machine Learning
Broadcasters
and
ASR engines learn from audio that is already
available, but much of the content that is used on
TV and film is not yet being utilized to train the
speech engines. Different from other broadcast
technologies, like transcoding software or rack
mount scopes where the software can’t learn
from the media content, machine learning for
94 • Broadcast Beat Magazine • www.broadcastbeat.com
Captioning for live and remote streaming
During the last few years, the Federal
Communications Commission (FCC) has strength-
ened its regulations on captioning for internet
video programming. All video media previously
aired on television must now be captioned when
repurposed for online distribution. And most
recently, FCC captioning rules have been extend-
ed to include live and near-live programming,
which must now be captioned for the internet
within 8-12 hours from the time of the original
broadcast.
For now, this rule only applies to content that has
appeared on television, but it is inevitable that
closed captioning workflows will have to funda-
mentally evolve to meet the needs of the ever-
increasing live streaming content that consumers
demand. Captioning online content sends a mes-
sage of inclusion that has a strong positive impact
on audience engagement and brand approval.
Reading captions also improves content com-
prehension, especially second-language viewers,
which means your message will come across
more clearly. Captioning means more viewers
and more engagement and a better understat-
ing of your message, regardless of whether
the video is playing on a mobile device or on a
monitor in a noisy airport terminal or busy sports
bar. Captioning also enhances online visibility as
search engines can index the text in your video
and improves data mining possibilities.
By using a cloud-based live restreaming service,
accurate captions can be married to the live
program in perfect sync, rather than always dis-
playing a few seconds behind the video, vastly
improving the experience for end users who rely
on captions. This is done by adjusting the video
latency slightly to account for the delay inherent
to all live closed captioning techniques.
The future is here
Cloud-based ASR solutions are already making
their way into modern closed captioning work-