16 - 19 October 2023 Hollywood California #MTS2023
16 - 19 October 2023 Hollywood California #MTS2023
UNINTERRUPTED CAPTIONING Harnessing Hybrid Cloud for Seamless Disaster Recovery Bill McLaughlin Chief Product Officer
16 - 19 October 2023 Hollywood California #MTS2023
CORE PROBLEM
How d o we man age th i s ch al l en ge?
1
2
Close d c aptioning for live programs is de live red straight to air, with limite d opportunity to QC or re ac t to proble ms
On most live US broadc ast programs, full c lose d c aptioning c ove rage is le gally re quired.
LI VE CLO SE D CAPTI O NI NG I S A U NI Q U E BRO ADCAST W O RKFLO W
Live closed captioning is nearly always delivered remotely into the studio
•
Low data rate enabled remote operation, even back in the modem days
•
In human workflows, stenos or voice reporters typically freelance
•
In automatic captioning, cloud has better results and more frequent updates
Live closed captioning is delivered direct to air with limited QC, because low latency between audio and captions is critical Live closed captioning is regulated for quality, latency, and completeness, arguably more than video or audio!
RI SKS Operator doesn't show up Software doesn't work Internet connection down
Audio not routed correctly Cloud service troubles
5
DE SPI TE RI SKS, FE W MI TI G ATI O N ME ASU RE S I N TYPI CAL W O RKFLO W •
Human-related risks require getting another human on the job, which takes minutes (at best)
•
Networking or software related risks may require restarts or even more extensive debugging, leaving program segments or more without captioning
•
Master control attentiveness to captioning may vary, meaning it could be several minutes before a problem is even recognized
16 - 19 October 2023 Hollywood California #MTS2023
A platform like iCap that provide s a c ommon c onne ction inte rface to automatic c aptioning and many human ve ndors provide s some mitigation, but de pe nde ncy on the c ore platform and basic inte rne t c onne ctivity re mains...
TYPI CAL U SE CASE S FO R O N-PRE MI SE S AU TO MATI C CAPTI O NI NG
Sensitive enterprise content (non-broadcast) Preference or mandate to avoid data export to cloud providers outweighs other considerations
Co s t Sens itivity in Hea vy -U s e, Sing le Cha nnel, Sta tic U s a g e Pa tter ns Cost efficiency of cloud on-demand services is least beneficial in static usage patterns that approach 24/7 usage
Disaster Recovery facilities in Broadcast Facility may be designed to rely on little or no off-premises or cloud connectivity to enable stripped down production approach for the duration of a weather disaster with a large impact on regional telecommunications
WHY IS LIVE AUTOMATIC CAPTIONING COMING FROM THE CLOUD? Quality / Accuracy Very large language models available in cloud with frequent updates to vocabulary and usage Scalability
•
•
Rapidly Improving Technology (and many competitive vendors) Cloud users are less likely to be committing to a fixed technology product that may appear outdated in as little as 1 year
Very attractive per-hour rates enabled through shared infrastructure
Outsourced Operations
No commitment to fixed costs in hardware and licensing, especially good for infrequent use or pop-up channels
Having vendor manage custom models allows switch from human captioning models without introducing new responsibilities to internal broadcast operations staff
9
16 - 19 October 2023 Hollywood California #MTS2023
OPPORTUNITY TO PAIR CLOUD & ON-PREM FOR MAXIMUM RESILIENCY Requirements: Fast as possible automatic detection of broadest possible range of problems
Probably wind up using preferred system 99.9% of time!
Rapid switching – definitely NO manual re-configuration of components
Ability to health check both systems even in large spans of time between outages
Use preferred system as much as possible to maximize quality and minimize costs
16 - 19 October 2023 Hollywood California #MTS2023
OPPORTUNITY TO PAIR CLOUD & ON-PREM FOR MAXIMUM RESILIENCY IN THE CLOUD (RUN BY AI-MEDIA)
Cloud Caption Data Source
CUSTOMER FACILITY
Video Channel Caption Encoders
OR
On-Prem Caption Data Backup Source
16 - 19 October 2023 Hollywood California #MTS2023
HYBRID CLOUD HAS ENTERED THE CHAT •
"Hybrid Cloud" is a general concept gaining popularity in many applications
•
Seeks to leverage some benefits of both on-prem and remote public cloud workflows
•
Examples: on-prem might be cheaper but less elastic, might be lower latency, might be more appropriate for DR and data protection needs in some cases
16 - 19 October 2023 Hollywood California #MTS2023
AI-MEDIA LEXI DR: HYBRID CLOUD APPROACH FOR CAPTIONING Individual caption encoders (SDI or IP video input) have fixed configuration identifying both public cloud resources (Lexi Cloud) and local resources (Lexi Local appliances)
Cloud Caption Data Source
Video Channel Caption Encoders
On-Prem Caption Data Backup Source
16 - 19 October 2023 Hollywood California #MTS2023
AI-MEDIA LEXI DR: HYBRID CLOUD APPROACH FOR CAPTIONING Live captioning to either system scheduled by: •
Calendar
•
HTTP API
•
Physical GPI
•
SCTE-35/SCTE-104 caption encoder channel
•
Detection of caption absence on video input
Activities on systems are synchronized up to the point of any communications incident
16 - 19 October 2023 Hollywood California #MTS2023
AI-MEDIA LEXI DR: HYBRID CLOUD APPROACH FOR CAPTIONING •
Caption encoding channel controls transition • From communicating with cloud Lexi platform • => To communicating with Lexi Local appliances if captions are not flowing from cloud
•
Agnostic to reason for failure: outbound network connectivity, cloud provider outage, misconfiguration
•
Very fast (as little as 15 seconds, only a few phrases of captioning)
•
Problems not solved: audio routing to caption encoding channel, failure of caption encoding channel, etc. • These issues best addressed by redundant caption encoding channels on site (A/B video chains)
16 - 19 October 2023 Hollywood California #MTS2023
Primary Mode
Cloud Caption Data Source
Video Channel Caption Encoders
On-Prem Caption Data Backup Source
Continuous Sync while facility link to cloud active Fast: within 15 Seconds
DR Mode
Cloud Caption Data Source
Video Channel Caption Encoders
On-Prem Caption Data Backup Source
Active Captioning when facility link timed out
16 - 19 October 2023 Hollywood California #MTS2023
SHARED ACCOUNT ARCHITECTURE FOR HYBRID CLOUD Lexi Local systems do NOT require continuous cloud connection • DOES use internet connection when available to sync custom vocabularies and other settings from the Lexi Cloud system Lexi Local system usage is also reported to Lexi Cloud and appears on one monthly bill with pure pay-as-you-go pricing • Break from most previous on-prem captioning models requiring fixed annual or multi-year license
16 - 19 October 2023 Hollywood California #MTS2023
SUMMARY OF BENEFITS OF HYBRID CLOUD LIVE CAPTIONING Combined reliability expected to approach 99.999% • •
Lexi Cloud real-world achieved reliability for typical broadcaster between 99.9% and 99.95% (Not counting scheduling / operational user error) Handovers to local capacity should eliminate nearly all remaining issues
Enjoying quality benefits of cloud-based automatic captioning on nearly all programming Retaining option to use human captioning through remote iCap on demand, unlike on a pure local appliance model Same pay-as-you-go pricing model as Lexi Cloud service
Conclusion / Q&A
Q/A Conclusion •
Live caption workflows are unique, but have followed a similar evolutionary path to many broadcast communications systems to cloud and software-focused
•
Hybrid cloud helps fill in the gaps for adding resiliency or plugging certain gaps (or saving costs) compared to total reliance on public cloud
•
AI-powered captioning accuracy is now competitive with traditional human transcription – this seamless DR shows how an automatic approach can be not just "as good" but "better"
GET IN TOUCH.
Website WWW.AI-MEDIA.TV
Email SALES@AI-MEDIA.TV
Phone + 1 516-293-7472