How will artificial intelligence and machine learning help drive efficiency throughout the M&E Industry? In this talk, John Motz will explore the future of machine learning and how it is enabling companies to automate manual workflows to become more efficient and more productive.
Speaker : John Motz, CTO, GrayMeta
4. Data now becoming
more available
Things have changed… with more change to come
Distracted
Bored
Leave
Expensive
Geographic Constraints
Siloed
Can be Expensive
Geographic Constraints
Human CapitalFacilities
Lots of data and history on this – good and bad
Scales well
Cost improves over time
Cloud Services
Gets smarter over time
Companies are building
platforms to ensure there is
commoditization and scale
Result: Decreased costs
Machine Learning
Cognitive / AI
5. Accelerating the opportunity at GrayMeta
Media & Entertainment
Compliance
Distribution
( IFE )
Distribution (New
Markets)
Distribution
(Regulatory)
Search, Recommendation,
Discovery
EPG, Apps New Tech – Bots
Ad Platforms and
Revenue
Generation
7. What happens when you use cloud translation
technology instead of highly trained people?
8. Watson’s latest feat of
94.1 percent accuracy
is fairly impressive —
though it should be noted, it’s still
well below the 98.5 percent
accuracy rate required by many
captioning companies.
A recent study, Global and
China Speech Recognition
Industry Report 2015-2020,
projected the global intelligent
voice market will grow from $6.21
billion in 2015 to $19.2 billion in
2020. In China alone, voice
recognition is expected to be a
nearly $3.8 billion market within
four years.
– David Ward
thejrc.com
Date: Nov 2016
It isn’t perfect, but is getting better everyday
9. Machine Translation for Subtitling: Large-Scale Evaluation, Jan 5, 2016
Productivity
“Taking the average productivity for
each translation pair, and considering
all machine translated files, filtered
and unfiltered, the gain in productivity
reached 38.2%.”
And
“Measures of productivity gain/loss
were also positive, with an overall
increase of nearly 40% in terms of
subtitles per minute.”
Accuracy
“Overall pretty good. Simple sentences
were usually perfect, but the machine
has problems when the sentence is
complicated.”
And
“Generally speaking, it was only in
very rare instances that the level of
translation generated was such that it
needed little or no editing at all.
Frequently, it was just easier to get rid
of everything and start from scratch.”
Perfection isn’t required to realize benefit
10. What are the opportunities?
Time Cost Quality Control
Notas do Editor
We have seen in 2016 and early signs already show 2017 wont be any different
You can be a creator, seller, buyers, channel, broadcaster, OTT or a Youtube channel
Creation, Distribution, Consumption of data and content is continuing to be in the rise / incline
More sellers and creators, its not all about theatrical release, short form is huge and growing
Buyers are growing
How good is good ?
The amount of time it is taking for content prep and compliance – be it airline, new int. markets – technology needs to be able to reduce this, even it is going to impact some of the existing services offerings or rate card line items
Sellers working direct with buyers ( IMF as an example ) – less transcode rev.
Technology Providers now able to offer the services. Microsoft Azure media Services, S2T
Buyers building platforms to control costs and quality and technology used
MAM, DAMS – newer ones on the market – Vidispine, Levels Beyond and more…. Built for the cloud vs yes we are in the cloud
There will be predators and there will be protectors … seen it in film scanning, seen it in new packages like UV, DPP, IMF
Things have changed for sure
There are the incumbents just like every industry, protecting existing market share, customers and revenue
Then there are the predators looking for some of the existing market share or to create new market share
There will be new platforms and systems and providers in the market place in the next 12-18 months..
There have been some great new entrants or great new technology in this space in recent years… Sfera did some great things early with the Cloud and AWS changing how things were done so much so seems many providers used them.. Now they are part of Deluxe – kudos to Morgan Fumi and team .. Zoo Digital is also another great one that at every NAB or IBC they have new versions of software, they want to push the envelope..
Maybe the future providers are the platforms doing it for themselves.. As long as there is the high dependency on human capital to get this work done then anyone can have a crack at it … these people tend to be free to work for whomever.. Maybe the future is every resource doing captioning etc has their own ID that goes with their content and the future is a open marketplace where jobs are issued and bid on..
So in 2018 who will have the control on how this work gets done? Will it be the same as it is now.. ??
To finish on a light note hopefully, we spent time looking at example of where this technology might not be ready for all use cases.
Came across this .. Not that service providers in the room use public internet services to translate but look what can happen if you just keep feeding the machine
There are plenty or articles, blogs and opinions out there
Data points seems to becoming more frequent
Improvements are gaining velocity
Body text: The latest on speech recognition - By David Ward http://thejcr.com/2016/11/29/the-latest-on-speech-recognition/
Some of you in this room were part of the EU paper – our review of it and we also asked many people to read and give their view is – not ready for prime time don’t stop everything you have already done
But its moving at such a pace in the right direction.
You just need to decide when you are going to start using or working with this technology and to what extend.
Its also ok to sit and wait
Body text: http://www.lrec-conf.org/proceedings/lrec2014/pdf/463_Paper.pdf
What are we looking to do leveraging Cloud and Machine Learning services
How can we use technology to connect data easier
Reduce time, cost
Sustain Quality or Improve Quality
Give more control ( when it is wanted )
Context set – while a number of our team come from the M&E space, much of our focus is outside of M&E also.
WHAT WE HAVE LOOKED AT
We have looked outside M&E at some industries that are bigger in spend but also are facing huge growth in video and content – this data type is pretty new for them unlike M&E
Law Enforcement, Public Safety, Govt and Healthcare, Fin Tech
Some of the initial areas we are looking at and testing with companies in the M&E space – Content Creators or Buyers / Service Providers
Compliance – object, word, content – safety = saves time, and cost tied to failed checked with broadcasters,
Leverage the speech to text – to improve search and rec engines and ad insertion for digital properties .. We are seeing interest to look at this tech for lower tier content not movies but talking head shows and news
What we are doing – How
We are working with a range of technologies some our own IP, ( open sourced ) and platforms which can leverage technology around ML,DL, Cognitive and AI to start – in sometimes very small ways to make a dent in some of these 4 key areas
There seems to be a good amount of conversation around S2T and then translation and language detection..
What is interesting is there are large companies out there working on things tied to the M&E but aren’t in the usual top 5-6 localization vendors