In Broadcast - January 2020 - 44

www.inbroadcast.com | Vol: 10 - Issue 1 | January 2020

InMedia

Accelerating and Automating
AI-Driven Closed Captioning
Russell Wise, SVP, Digital Nirvana, considers the options for AI-boosted
closed-caption generation in existing production platforms...

rtificial intelligence (AI) and
machine learning (ML) processing
in the cloud today offers impressive
accuracy, scalability, and trainability
across a variety of applications useful
to broadcasters. To benefit fully from
these services, however, broadcasters
must be able to integrate AI-driven
and ML-driven microservices into the
on-premises workflows that already
support day-to-day operations.
While many broadcasters do take
advantage of cloud-based speech-totext (STT) services for closed-caption
generation and subtitling, most find
the process too time-consuming and
resource-intensive to perform in
house. The process typically involves
creating and sending proxy versions of
content to the cloud for STT processing,
retrieving and then integrating
caption data into video, correcting
and adjusting captions to suit the
requirement of target distribution
platforms, and exporting the final
captioned output in the appropriate
format(s).
Outsourced Closed-Caption
Generation
By
outsourcing
closed-caption
generation,
broadcasters
can
eliminate some of these manual
processes. Even better, though, in
terms of time and cost savings, is the
option of integrating, automating,
and orchestrating closed-captioning
as part of the overall broadcast
workflow. Broadcasters can leverage
the intelligence and flexibility of
sophisticated STT engines in the cloud
while working within an intuitive
interface with all the functions
essential to delivering compliant,
conforming high-quality captions.
Although the technology behind this
model is sophisticated, the captioning
workflow is simple. Files transferred
via API or portal from a MAM, PAM,
or storage system (cloud or local)
into a captioning watch folder are
automatically transcoded into the
proxy format best suited to the host
cloud platform and STT processing
engine. Because STT engines today
have been trained on billions of

Digital Nirvana offers its
Trance enterprise-level solution

words and on thousands of hours of
spoken data representing a wide array
of accents, the captions generated
are remarkably accurate. Speaker
detection and spell-check functions
add to accuracy and further reduce
the time needed later for quality
assurance.
Paired with timecode data, a
transcript is delivered back to the onpremise system and stored along with
other critical metadata - all accessible
through the operator's MAM. Within
the user interface, time-synched video
and captions are shown side by side,
along with tools for editing text and
adding visual cues, music tags, and
speaker tags. Dictionaries, transcripts,
rosters, and other material specific to
a media asset can be ingested by the
system and applied to the transcript
to increase accuracy and help ensure
unique names and terms are correct.
Built-in automatic error detection
identifies other possible issues,
helping the operator speed through
the correction process. Even when
working with sports and other shows
that include slang and specialized
jargon, the operator can quickly go
from marker to marker or line by line
to correct or enhance the copy. At the
same time, the operator can examine
and correct metadata associated with

the video, whether originally delivered
with the media asset or added during
STT processing (or processing by
another cloud-based cognitive engine).
Platform Style Preset
Every distribution platform broadcast and OTT - has a style guide
in terms of how closed captions should
be presented. In a sense, the style
guide is a brand book, but for closed
captioning. Content that doesn't
align with the target platform's style
guide will be rejected. By configuring
presets for each platform, broadcasters
can automate and accelerate the
formatting of captions according to
the target platform's style. Number of
characters, number of lines, caption
placement, and other elements can be
specified in each preset.
The
corrected
transcript
is
automatically converted into timesynced captions in accordance with
policies defined in the appropriate
preset. The resulting captions are
displayed in a captioner window for
final comparison to video and review.
Once the captioner has made any
further changes, the file is used to
generate the multiple output formats
required for distribution.
Because they are leveraging cloudbased microservices, broadcasters also

Trance has a cloud-based architecture

have the option of using a translation
engine to automatically generate
multilingual transcripts. Once again,
the transcript can be automatically
generated and made available to the
operator in a dual-pane windowed
display for review and correction.
Spell-check and error alerts reflect the
language of each source transcript.
Trance Captioning Process
By building the closed captioning
process into their existing operations,
broadcasters can decrease their
costs and increase the productivity
of in-house teams by orders of
magnitude. They can guarantee
highly accurate closed captions with
fast turnarounds. They can deliver
accurate and compliant content for
publishing worldwide. Orchestration
of the whole process delivers further
gains by centralising workflows such
as job allocation, caption edits, role
assignment, report generation, and
status monitoring for streamlined
project management.
Platforms exist today that bring
these capabilities to the broadcaster.
Once such solution is Trance from
Digital Nirvana. Built on a cloudbased architecture, this enterpriselevel solution offers an integrated
and automated speech-to-text (STT)
engine, with an easy-to-use interface
and intuitive orchestration tools that
bring significant new efficiencies to
captioning and subtitling workflows.
As a platform capable of supporting
multiple applications, Trance also
gives broadcasters the opportunity to
incorporate metadata enhancement
and content classification into the
overall workflow, thereby adding value
to every piece of content.
AI platforms capable of hosting
multiple cloud-based microservices
- closed captioning / subtitling and
others - give broadcasters tremendous
flexibility in processing and delivering
the increasing volume of content that
publishers and consumers demand.
Bringing the benefits of cloud-based
processing in-house and using
orchestration to boost efficiency,
broadcasters can handle changes in
scale and in distributor requirements
with agility - and within the familiar
media workflows that drive their
business.

http://www.inbroadcast.com https://digital-nirvana.com/trance/ https://digital-nirvana.com/trance/

In Broadcast - January 2020

Table of Contents for the Digital Edition of In Broadcast - January 2020