Ref. Ares(2017)4615423 - 21/09/2017
Ref. Ares(2017)4763782 - 29/09/2017
From:
(CNECT)
Sent: Friday, January 20, 2017 12:11 PM
To:
(CNECT)
Cc:
(CNECT);
(CNECT);
(CNECT);
(CNECT)
Subject: call with Audible magic
Personal Data
,
A quick summary for you to follow the discussion we had with Audible Magic on
Wednesday (18.01).
is speaking to INA today and we may try to contact also
other technology providers to get more info.
***
Slide 3 (chart of effective technologies) – it includes estimates made by AM (average
costs and effectiveness), mainly based on their own data.
Effectiveness of a technology depends on the type of content:
-
File name (metadata) could be a first indication of what the file can contain, eg if
the name of a song or artist is given in metadata
-
Fingerprints: the most efficient in general terms but the levels of actual efficiency
will vary, eg it can be very efficient for sound recordings and not so much for
recognising compositions.
-
File hash: can be effective for enforcement purposes on P2P networks , not so
much on UGC networks as each hash is unique which means that each use would
create a new hash even if it may be the same content , ie there are as many hash
files as ripping of a same content: as a result, hash file technology is good to
"validate" that a particular file encapsulates a specific content, but not to identify
and block all files that encapsulate such content )
-
Watermark technology is the least documented
At the same time, effectiveness assumes that the process can be measured. But it is
not how the market works today – there are too many files out there. What needs to
be achieved is a balance between the expectations of RH and the affordability of a
technology for service providers.
With regard to cases where a service makes available different types of content and
would need to check all these, currently there is no service provider with a blended
technology that could deal at one go with different contents. Depending on the
characteristics of a service, it may therefore need to acquire different types of
technologies (which would entail additional costs).Technically, it is feasible to apply
different technologies to a particular upload even subsequently or in parallel
(checking is very quick).
With regard to
costs per transaction, the identification of a large file may involve
more than one transaction. The longer the file, the more numerous the transactions
(request sent to the system). So one transaction does not always equal one upload.
Pricing is based on the number of transactions carried out per month per slice (eg
5000-100 000 transactions/month; 100 000 – 1 m transactions/month; 1million to
100 million, and the last tier is beyond 100 million transactions per month) SME
cases studies mentioned in their presentation (Spinnup and VNG: under 4000/5000
uploads per month) so this would fall within the first tier which is considered small.
Of course, pricing will decrease with the increase in the volume of transactions
checked.
As to actual practices by platforms/their clients, they don't know whether all user
uploaded content is checked or only a selection. But it is difficult for a platform to
discriminate between uploads, so probably there is no picking and choosing. AM
provides the customer with their software which creates a fingerprint of an upload
and then it is sent (as a separate workflow) to the central database of AM, to check
matches. It is for the customers to decide what is checked (possibility to
discriminate) or, in case of a longer file, to send for identification only a portion of
the file. AM has no way to control this. The checks may also be made at a different
time than at the moment of upload, e.g. some services send for identification only
the content with a large number of views, or only once it gets viewed for the first
time.
We asked them why Soundcloud uses 2 technologies (their own and AM). They
explained that AM is used generally ex ante, whereas they probably use their own
technology for checking post-upload.
Vis a vis users, pre-emption (content is blocked before upload) provides better user
experience than takedown (ex post).
Document Outline
