MultiChoice is Africa's leading video
entertainment company
and we service around 20 million customers
across 50 countries.
Captioning is very important
to us at MultiChoice,
particularly
to allow accessibility of our content.
And this is primarily for making content
more accessible to our customers
that may have hearing difficulties
for one.
And the second purposes is around
repurposing our content,
making our content more accessible
outside of South  Africa, particularly in Africa.
The brief that was provided from
MultiChoice was actually quite simple.
It was help us reduce the time to edit
content firstly.
Secondly, help us get content
to more of our users by taking languages
like English as an example,
and translating it to other languages
that are quite unique to Africa
as example Zulu or Kiswahili and vice versa.
Taking content
from international countries
and again translating that
into local content or local languages
should I say, so that we can sort of give
content out there to a wider audience.
So we produce
a large volume of local content
and manually capturing this content is
proving to be very difficult
due to the
volumes and due to the strict
turnaround times needed for broadcast.
Well we were introduced to AI,
and it helped us
from creating subtitles from scratch.
If you don't have supporting documents
for the content.
So in that way it made it simpler
because it was faster,
after comparing the times of starting
from scratch and also then using AI.
So I'd say this made it much easier.
So when we started off with this project,
there were basically two major challenges.
The first challenge was how to run some of
this international content on prem.
So the second challenge was
how to handle the various accents
and dialects within South Africa.
The English spoken here
has peculiar accents and so forth,
and the models needed to
perform well for this type of
content.
And from that, essentially
what we did was we created a virtual team
because this is not something
you fix with one person,
and it's not just technology
you can drop down and it works.
It has to understand content,
it has to understand languages.
MultiChoice is the first customer to use
South African English in their content.
They were the first customer
to look at Zulu or Afrikaans, et cetera.
So what we did was engage
our engineering teams
from the Microsoft Speech Services
and started bringing them
into the discussion into the fold
and created a program around
subtitling.
For this application,
We basically used a hub and spoke
kind of a model where the hub is the place
where you have all the common services,
which gets shared across use cases.
And each of the spokes
basically represent a new use case
which comes in, gets
added on to this framework.
So subtitling is one of the spoke
translation is another spoke
which gets added to the main project hub.
So having this
type of collaboration on the ground
with MultiChoice
getting real ground truth information
in a real location
and providing that feedback
back into the organisation really helps
because we can fine tune these models,
we can continue to improve upon them.
Working collaboratively over
six months really helped us
accelerate this use case and also feedback
some of the findings
that we had, back to Microsoft
so they can also improve on their models.
The success of this service relies
on the content
that you give to the service
to train itself.
So what's really important
and specifically in this project,
what makes it unique
is that we were able to get a variety
of different media sources
through MultiChoice,
whether it was local
content or international,
and why that's important
is the more content
you give to the service,
the speech service on Azure,
the more it learns and the more
you tell it where it makes mistakes,
the better it gets.
I think one of the most unique features
that we got with the Microsoft stack is
the ability to process content
within our buildings,  so on-premise
because we do have sensitive content
that we cannot send out of
the MultiChoice building.
So I think with the Microsoft offering,
we were able to leverage this
and also process this content
within our own environment.
You get some of the services
managed by Microsoft
that is actually very great for us
because the speech to text model,
we don't need to worry about
how the model is being trained
how it's getting improved.
Microsoft handles that for us.
So what we as a user
just have to integrate with their API
and keep using it.
The accuracy, I think can be measured
in different things.
Firstly in timing, it's
highly accurate, right.
So we get it on...
I think I personally give it between
90% and 95%.
By collaborating with Microsoft,
it really allows us to accelerate
our AI use cases
and be the leaders in Africa
with this type of technology.
Our company CEO, Satya Nadella
has been talking about the next
AI revolution happening in Africa.
And we are seeing it's happening
right before our eyes
with these capabilities on languages
and cultural considerations,
Addressing issues that affect Africans
and for Africa, it is the way to go,
and Microsoft
is really pushing that agenda forward.