A Buyer-Seller Protocol with Watermarking for Cloud Streaming

(1)

IN

DEGREE PROJECT ENGINEERING AND ECONOMICS, FIRST CYCLE, 15 CREDITS

STOCKHOLM SWEDEN 2017,

KTH ROYAL INSTITUTE OF TECHNOLOGY

A Buyer-Seller Protocol with Watermarking for Cloud Streaming

Towards an Ecosystem for Media Streaming

Johan Björklund

johbjo09@kth.se

June 2017

(2)

Abstract

This work shares purpose with new directions in the philosophy of intellectual property, where self- expression and participation in the creation of culture are seen as key ingredients of human wellbeing and autonomy. A technical solution is explored that enables major labels and independent creators to publish music with equal reach to audiences, without the need for trusted third parties.

The principal contribution is a buyer-seller protocol in the setting of untrusted service providers and a blockchain ledger. In the envisioned scenario, files are streamed from untrusted providers to end users. Encryption and watermarking, rather than obfuscation, are used to protect against and disincentivize piracy. Watermarking bit rate, storage and communication overheads, and encryption performance are key parameters. Subscription and ad-supported pricing models are discussed.

(3)

Sammanfattning

Detta arbete delar syfte med nya idéer inom immaterialrättsfilosofi, där yttrande och delaktighet i skapandet av kultur ses som nödvändiga ingredienser för mänskligt välbefinnande och självständighet.

Rapporten utforskar en teknisk lösning som möjliggör för stora mediebolag och oberoende artister att publicera musik med samma åtkomst till publik, utan behov av betrodda tredjeparter.

Det huvudsakliga bidraget är ett köpar-säljar-protokoll för en miljö med obetrodda service providers och en blockchain ledger. I det föreställda sceneriot skickas filer från obetrodda parter till slutan- vändare. Kryptering och vattenmärkning, snarare än obfuskering, används för att skydda mot och avskräcka från piratkopiering. Vattenmärkningens bithastighet, lagrings- och kommunikations-behov, samt krypteringens prestanda är avgörande parametrar. Subscription och ad-supported som prismod- eller diskuteras.

(4)

1. Introduction

At the end 1990s, the music industry was the first of the media sectors to be exposed to the new forces brought on by advances in telecommunication and computers. The growth of the internet both in bandwidth and number of users was enabling mass distribution and dissemination of recorded works such as songs and movies. The record industry was forced into a disruptive transformation [Dol11].

On-demand multimedia, or streaming, has now become the dominant segment of the record industry.

Historically, distribution of recordings was achieved through either broadcasting or physical distribution of recorded media such as discs. These mechanisms afforded industry actors (publishers, record companies, broadcasters) natural entry barriers and dominant strategies which are no longer in effect.

These barriers, however, together with copyright law, made it possible for industry actors (publishers and record companies) to gather large catalogs of copyrights, and trading in these as commodities has become the core business of these companies.

Consolidation among publishers and record companies has resulted in a large portion of the copyrights of popular works being concentrated in a handful of companies, now simply called media com- panies, the largest of which are referred to as the majors. These catalogs grant them dominating strategies against streaming companies who need to obtain distribution licenses. This in turn causes distortions in what would otherwise fulfill the conditions for an open ecosystem and a free market for streaming services, composers, artists, and listeners.

Historically, industry actors such as publishers and record companies provided several services to composers and artists that would otherwise have been difficult to obtain. Recording studios and equipment have historically been capital intensive to build and maintain. Artists who signed record contracts were given access to studios and record producers. Records were pressed in the companies’

plants. In 1997 the majors held over 80% of the market [Dol11], each company controlled the entire value chain - from talent selection, studio provision, to record pressing. The contracts generally stipulated a royalty on proceeds from sales of licenses of the recordings, while the record companies were assigned the copyright of the recordings. Distributing records required access to distribution channels, and channels to promote plays radio broadcast serving as advertising for the songs. Because these mechanism exhibit economies of scale with respect to the size of the copyright catalog, there were incentives for composers and artists to sign contracts with record companies.

Growth of the internet and has weakened some of these incentives. Personal computers and studio emulation software have enabled artists to produce quality song recordings at low cost. The important production factor is now creativity and talent, both of which flourishes and readily self-assembles through social networks. Collaborations can self organize through social networks. Distribution on the internet is essentially free. Advertising of songs is achieved through social networks and recommendation algorithms. These factors have enabled global music careers to be launched from hobby studios.

These factors have conspired to dispossess the traditional music industry of its protective barriers to entry, and it is now relying solely on its core commodity - copyrights. Streaming platforms, functioning as markets for music, need to include the historic catalogs in order to attract audiences that are big enough to sustain them. Majors can still dominate the market by stipulating conditions in licensing contracts of their catalogs. Independent artists, who want to use streaming platforms for promotion and distribution, are put at a disadvantage relative to the majors.

(7)

Purpose

A technological solution to these problems would give each participant equal access to the streaming market. Imagine an ecosystem where parties were free to participate in any of the following roles:

Content owners Composers, artists and publishers would publish their songs in the system. They would decide the terms and prices for users listens to them. Owners would compensated when songs are streamed.

Service providers The eco system would depend on network bandwidth and storage capacity. The allocation of these resources should be compensated by users in accordance with their specific utilization; such that popular songs consume more resources, and thus should cover more of the costs.

Service providers would be compensated for each song they stream, and they would be independently responsible for identifying high demand songs, and offering capacity for those songs. This requires that demand information is available to providers, and that they are free to offer or deny capacity to any song they want.

Thus, capacity in the system would be regulated, on the level of individual songs, by a supply and demand equation. When this is realized, a solution to the resource problem of the long tail is also achieved, in the sense that high demand songs will “pay for themselves” whereas low-demand songs will not be profitable to host.

Payment providers Participants would be able to receive and provide payments for services through more than one payment provider. This would make different payment schemes available to users with different needs. If payments is a separate function in the eco system, business models such as different payment schemes become possible. Some users prefer monthly subscriptions, whereas some users prefer to pay only for songs they listen to as they go, others might prefer prepaid monthly quotas.

The business models implemented to enable these schemes are closer to financial services. Free entry would enable competition on payment plans.

Contracts Access to the ecosystem would be defined by protocols, not by negotiations of contracts.

This would defuse attempts at collusion, or one actor gaining unfair advantage by controlling many links the value chain. The scheme would enforce copyrights through both encryption, incentives and disincentives.

1.1. Research question

This work centers around the question of the feasibility of an ecosystem such as described so far. At the center of such an ecosystem is a communication protocol that must fulfill certain requirements. To articulate these requirements, we must first understand how social incentives and business objectives can be coded into functional requirements, and then realized with cryptographic methods. Specifically, this work aims to deliver answers to the following questions:

• What are the requirements of such a protocol?

• What known and available methods can be used in the construction of such a protocol? By known and available is meant methods that are implementable, and not only theorized.

• How can these methods be applied, and what compromises are there?

The result of this work takes the form of a communication protocol. The properties and parameters of this protocol give qualitative answers to these questions.

(8)

1.2. Methodology and scope

Starting with a survey of related attempts in the field, with the aim to understand the conditions that allowed them to succeed or caused them to fail. A background on information economics, network externalities, and intellectual property helps provide a setting for the problem. Conclusions from the background study are compiled into an outline of an ecosystem which critically depends on encryption, fingerprinting, key exchange, and smart contracts.

The scope of this work is then to investigate a protocol with a special key exchange that is implementable with fingerprinting and contracts on a blockchain ledger.

Audio watermarking and blockchain ledgers each have multiple candidate implementations serving varied purposes, and improvements in specific techniques may in the future make them feasible in wider settings. The methodology in this work is therefore to center on the key exchange scheme, and investigate how the parameters affect realizability. Watermarking, fingerprinting, and blockchain ledgers are therefore studied to understand requirements and parameters.

A literature survey investigates the current state of audio watermarking, buyer-seller watermarking protocols, fingerprinting, and decentralized payment transactions with the aim to:

1. Find abstractions and primitives that can be applied such that

2. the security of our scheme relies on the provable security of the primitives.

To understand collusion resistance in fingerprinting codes, we look at Boneh and Shaw’s scheme which uses error correcting inner codes in combination with random outer codes. Tardos develops fully probabilistic codes and achieves an enormous improvement in code length over Boneh and Shaw.

Tardos’ fingerprint code length is a key parameter in our scheme.

In a sequence of papers on buyer-seller protocols, starting with [MW01], the so-called customer’s rights problem is laid out. The customer’s rights problem is seemingly unsolvable when using probabilistic Tardos codes. While solutions have been proposed, it is out of scope of our key exchange scheme. This limitation of scope is justified in section 3.2.2.

A survey of literature on audio watermarking and data hiding results in an overview of techniques with parameters that are adaptable for our scheme. Embedding bit rate is critical to collusion resistance, and it is incorporated as a parameter in our scheme.

Surveying papers on blockchain protocols, we find in [PSS16; PS16] straightforward abstractions of blockchain transactions that we incorporate in our scheme. While the abstractions we use are simplifications, they are intended to be realizable with state-of-the-art blockchain implementations.

We then define a communication protocol between three parties that enables transfer of fingerprinted files in return for value exchanged on a blockchain ledger, incorporating relevant parameters and showing relations between them.

A proof of correctness of our protocol is given, but we defer security analysis and proofs to some other time and place.

Report audience

Blockchain technologies have intrigued many in and around the music industry. Among established actors, it is primarily being envisioned as a solution to the problem of cataloging and matching songs to rights holders, composers, and artists by providing a shared transactional store of this information.

So called blockchain ledgers can also enable decentralized transactions of value, of which cryptocur- rencies such as Bitcoin are examples. In the setting of music distribution, they enable payments to flow directly from consumers to producers without intermediaries. Blockchains used in this way have interested primarily industry newcomers such as start-ups and artists. What has been missing are concepts that can work under realistic assumptions.

(9)

1.3. Related work

Broadcast encryption

Broadcasting is characterized by one signal reaching many receivers, and broadcasting to paying subscribers poses the problem of preventing non-subscribers among receivers to disseminate the content.

This can be achieved by encrypting the broadcast signal and only sharing the key with confirmed subscribers. If, however, the key is stolen or shared by one of the subscribers to pirates, the problem of key distribution and revoking is raised.

Broadcast encryption addresses the problem of key distribution and revoking in a setting with one signal and many receivers. Traitor tracing is the problem of, upon discovery of a pirated key, finding the user or users who are responsible for sharing the key.

As streaming is not analogous to broadcasting, broadcast encryption is only faintly relevant to our study.

To get a feel for problems in broadcast encryption, we take a quick look at key hierarchies. Key distribution and revoking can be achieved by generating keys in a binary tree structure, where each node is a key. Each leaf node is assigned to a subscriber, and the subscriber knows all keys along the path from the subscriber’s node to the root. Broadcasts are encrypted with the root key. Revoking one subscriber then means replacing all the keys he knows. This can be accomplished by encrypting update messages with keys “sibling” to the nodes along the subscriber’s path, so that everyone except the revoked subscriber can read them. The binary tree reduces the number of update messages log₂N in number of subscribers.

Broadcast encryption accomplishes identical plaintexts for all subscribers, which is not applicable.

Fingerprinting and watermarking

Fingerprinting is a scheme to disincentivize piracy by embedding a signature (“watermark”) into a sold good in order to uniquely identify the original buyer or user. Thus, if a pirated copy is found the responsible user can be identified by decoding the watermark.

Fingerprinting of data is first introduced as a mathematical problem by Wagner in [Wag83], where an example of statistical fingerprinting is given.

Boneh and Shaw [BS98] describes a fingerprinting scheme in coding theory, gives definitions for security against framing and collusions. It is shown that a fingerprinting scheme can not be completely safe against attackers, and it is proven that a fingerprinting scheme can be secure against collusions and framing up to certain probabilities, given assumptions about the attackers.

Tardos [Tar08] gives a scheme that improves on Boneh and Shaw by using a random binary code, allowing smaller codes for equivalent security.

Buyer-seller watermarking

Memon and Wong [MW01] proposes a scheme of protocols where the buyer is issued a random fingerprint by a trusted central authority. The fingerprint is encrypted with the buyer’s public key. The seller then embeds both the buyer’s fingerprint, and the seller’s own random fingerprint into the file - both encrypted with the buyer’s public key, and saving tuple of the buyer’s identity and encrypted fingerprints in a table. The file is delivered encrypted with buyer’s public key. Through homomorphic properties of the encryption scheme, the clear watermark will be embedded in the decrypted file. If the seller finds a pirated copy, it can extract the fingerprint from the suspected file, and look up the most likely buyer/pirate from the table. A buyer can prove his innocence by decrypting the fingerprint (stored with seller) and showing it is not present in the pirated file.

Lei, Yu, Tsai, and Chan [Lei+04] address the customer’s right problem, the unbinding problem (the seller can extract the watermark in recovered unencrypted files from the buyer and embed into other files). Their protocol relies on a trusted third party, but they propose solutions to the unbinding problem.

Shao [Sha07] addressed anonymity and reliance on trusted third partys.

(10)

Williams et al [Wil+09] investigated both Lei and Shao’s protocol using a type of formal analysis.

Both their solutions were found to be vulnerable to unbinding attacks. Specifically, once sellers had apprehended pirated copies, they could extract watermarks and embed in any other content.

Terelius [Ter13] investigated buyer-seller watermarking in the setting of a peer-to-peer network, and the problem of transferring the ownership of watermarked files, but ignored the problem of collusion.

The protocols described so far rely on short fingerprinting codes and/or embedding through homomorphic encryption; neither of these are realistic for audio data.

Kiyaias et al [Kia+15] describe an asymmetric fingerprinting scheme using Tardos-code. An asymmetric fingerprinting scheme solves the customer’s rights problem by letting the file distributor be oblivious to the fingerprint code. In [Kia+15], this is achieved by a crypto primitive called oblivious transfer.

Obfuscation

In [Bon+15], Boneh addresses problems with hosting services on untrusted clouds, which is tangentially related to our study. Specifically, he uses Indistinguishability Obfuscation, a pimitive that “shrouds”

programs such that their algorithms are hidden from the machine executing them.

1.4. Contribution

Previous work on buyer-seller watermarking schemes focus on the customer’s rights and the related unbinding problem - scenarios in which distributors, because they know the fingerprint codes, are able to distribute watermarked files and thus frame legitimate buyers. There are two assumptions behind this focus:

1. That a particular watermarking scheme is sufficiently reliable to make accusations viable in courts of law, thus enabling distributors to profit from false accusations.

2. That distributors may have access to plaintext content.

In the setting of streaming of music files from untrusted cloud providers both of these assumptions fail. Fingerprinting using audio data hiding in music have properties that make accusations weak in courts of law. Punitive actions are likely to be done outside of courts, and not be financially beneficial to accusers unless in very special circumstances.

The assumptions behind our scheme are instead:

1. That an encryption and fingerprinting scheme can be so inconvenient to circumvent that it is not cost effective for anyone to even try.

2. Distributors are untrusted and must not be able to exploit their role. Thus they may not have plaintext content.

With these assumptions, we give a buyer-seller watermarking protocol in the setting of untrusted cloud hosting providers and a blockchain ledger, that has some realistic chances of being implementable.

1.5. Structure of this report

Chapter 2 investigates the background of the current state of music distribution, covering intellectual property, the economics of music, and current streaming businesses.

Chapter 3 starts with a compilation of the requirements, and investigates the three main components of the protocol; fingerprinting, key-exchange and encryption, and smart contracts on a shared ledger.

Chapter 4 contains an ideal definition of the protocol, an implementation with discrete logarithms, and some notes on implementing with Curve25519 including a small performance test.

Chapter 5 summarizes the results and gives a discussion on pricing models, reasonable security, and feasibility.

(11)

Appendix A is a review of topics mainly from number theory and cryptography, and is referenced throughout the report.

Introduction

Theory Preliminaries

Economics of Music

Streaming and Downloads

A new Ecosystem

Fingerprinting

Key-exchange and Encryption

Smart Contracts

A Buyer-Seller Protocol

Conclusions

Optional reading Chapter 1

Appendix A

Background Chapter 2

Chapter 3

Results

Chapter 4

Chapter 5

(12)

2. Music distribution

2.1. The economics of music

Music is a form information. Characteristics of both its production and consumption make information into a special class of economic good. It is kind-of excludable, meaning access to it can be limited, and non-rival meaning it can be simultaneously consumed by many. These characteristics make it a type of public good. It is non-excludable in the sense that once information has been disseminated, it can not be revoked; a court of law can not order someone to forget the contents of an email. In the modern economy, supply of goods is governed by information (software and formulas in industrial processes), and demand is governed by information through branding and marketing.

2.1.1. Information economies

There is a distinction between creating and (re)producing information. Information can be enhanced and refined through production, but original information is created by minds independently of how it is recorded and reproduced. There are three important factor inputs for creating information:

• human creativity and inspiration

• tools; to help amplify creativity, and to record

• information; for context and inspiration

The relative importance of these factors depend on their relative availability.

The industrial information economy

Before computers, the economics of information production was governed by industrial processes in broadcasting networks, printing presses, and physical distribution. We will refer to these simply as industrial reproduction platforms. If the releases were large enough, larger platforms (by economies of scale) meant lower marginal costs per copy. This meant that operators of printing presses, through publishers, were incentivized to produce fewer releases in larger volumes, which meant incentives to invest more in each release. In the case of music, this meant bigger and better equipped studios, with more technicians and professional musicians.

Bigger reproduction platforms and bigger studios meant larger capital investments and increased barriers to entry. This meant fewer actors and monopsony-like conditions. Creatives - composers and musicians - had few clients to sell their services to. Industrial mass production gives uniformity.

Broadcasting, by example, only provides a limited number of channels. Maximizing the return and utilization of the reproduction platforms meant standardizing the inputs. The creatives, creators of the original inputs, became professionalized.

The need for capital investments and the ensuing need for professionalism dominated the need for creativity, which was in relative oversupply.

The networked information economy

With increasing power of personal computers, consumer electronics, and wider availability of edit- ing software in the early 2000’s, quality audio/visual production became affordable for independent creatives. Advancements in communication bandwidth and spread of internet devices enabled the

(13)

distribution of their works to wide audiences. Social networking sites and communities made it simple for audiences to find new artists.

In this setting, the creatives are not motivated by terms of professional contracts. The question of whether creativity can be motivated by contracts is touched on in [Ben06, ch. 4], from which the following quote lends some intuition on the difference between cultural and professional motivation:

“If you leave a fifty-dollar check on the table at the end of a dinner party at a friend’s house, you do not increase the probability that you will be invited again.“

Professionalism is necessary when people need to fulfill contractual obligations. Human creativity is the process that outputs originality. Improvements in the tools are making professional technicians luxuries. The tools available to creators are used to both amplify creativity and record works.

Benefit Public Domain Intrafirm Barter/Sharing

Rights based

Money from licensing.

Romantic maximizers Authors/composers sell to

publishers.

Mickey Disney reuses content.

RCA

A group of companies hold a pool of blocking

patents.

Non-exclusion-market Benefit from information, but not through selling licenses.

Scholarly lawyers Consultants present research to get clients.

Artists play music to grow audience.

Know-how More efficient processes

through internal information.

Learning networks Information shared in

engineering societies, mailing lists.

Non-exclusion, non-market.

Joe Einstein Gives away information

for free in return for status, benefits to reputation, or other

motivations.

Los Alamos In-house information to

provide valuable public goods, to gain government funding

Limited sharing networks Share among small group

to receive feedback.

Share with special licenses, like copyleft.

Table 2.1.: Benkler’s Ideal-Type Information Production Strategies. [Ben06, ch. 2]

The growth of the volume of material on sites such as YouTube and SoundCloud is explained by scholarly lawyers, Joe Einsteins, and limited sharing networks, using new software tools to create and publish their works.

The marginal cost of reproduction is approaching zero, independent of the size of the release.

2.1.2. The popular economy and music

The peculiarities of the demand for music characterizes it as a cultural good. A consumer’s preference for a cultural good is not decided by its utility, but rather by the meaning, pleasure and social identity for the consumer [Fis87]. The demand for a piece of music depends on the consumer’s subjective relation with the piece, and of cultural trends and tastes, which in turn depend on what exposure the piece has had. Demand, in a way, depends on previously supplied quantity. This is in contrast to other forms of information, such as news and fashion, which lose their value with time.

This implies a demand curve for music that is more dependent on previous exposure than price. It is exemplified in the distribution of tastes among age groups, where “classics” are more popular in higher age groups than younger groups, who have had less exposure. It explains phenomenons such as charity concerts, where artists who would otherwise charge substantial ticket prices, are able to perform without cost in return for exposure.

Music, being a cultural good, carries meaning and identity. Music of particular genres and time periods adds meaning and identity to occasions and people.

The challenge of music promotion and marketing is primarily to achieve exposure in culturally meaningful settings.

(14)

Records, such as vinyl discs and CDs, were sold at prices far above marginal cost, and prices were rarely differentiated. Each record held a finite number of songs, and was commonly an album from one artist. The number of records a customer could afford was limited, and average customers bought records based on exposure. Exposure of music through broadcasting was likely pareto distributed, which meant that the few records an average customer bought were picked from a pareto distribution.

This meant that artists needed to overcome a threshold of exposure for their records to be profitable.

In the 1980’s, 12 out of 13 records failed to make a profit [Fis87].

Why record companies choose not to differentiate prices can be explained by “mainstream culture”

and monopoly optimization. Since record companies had monopoly on the 10% of records that (by chance) reached critical mass, prices were set to optimize total revenues. It was probably efficient that records unlikely to reach critical mass failed quickly. This put niche artists at an unfortunate disadvantage.

Payola When radio was an important music platform, record companies would pay radio stations to play their records to get exposure of songs and artists they were promoting. This practice was referred to as payola.

2.1.3. Intellectual property and copyright law

In economic theory, welfare is maximized when the price of a good equals its marginal cost. The marginal cost of reproducing information is zero. One can argue that it would be unfair to creators if information was free, and that information would be under-produced. Arguments debating intellectual property usually belong to one of four branches.

In the first branch, information is seen as a public good which is under-produced in competitive markets. Therefore, governments should intervene by granting creators temporary monopolies (such as patents or copyrights), or outright subsidizing creators. This gives incentives for creators to think up ideas beneficial to society. Counter arguments include deadweight loss, frictions, and inefficiencies.

Fashion, furniture (not copyrightable) refute some of these arguments by example.

In the second branch, ideas are seen as “fruits of labor” which should be the property of its creator just as a material thing is property of its creator [Hug88]. Ideas might not be the deliberate goal of labor, but the byproduct of a (possibly wasted) effort pursuing something else.

In the third branch, ideas are thought to be extensions of the author’s identity and so the author should be granted control over these.

In the fourth branch of theories, information is seen as the “language of culture”. A rich cultural language is beneficial to all. Self expression, developing artistic competencies, building relations and connections around cultural pieces, are important ingredients for human wellbeing and “the good life”.

Semiotic democracy is the wide participation by people in the creation of cultural meaning [Fis01, p.

34], as opposed to passive consumption. According to this branch of intellectual property theory, laws should incentivize wide participation and enrichment of culture.

Copyright legislation The Berne convention (1886), by now adopted by most countries, lays out the foundations of current copyright law. It states that copyrights are granted to the creator automatically upon creation.

Copyright assignments In American copyright law, from 1909 to 1978, authors were required place the copyright mark on their works for them to be protected. Copyrights were granted to the author for 28 years, after which they could be renewed for a further 67 years. Authors were required to register their copyrights to be valid for renewal and in infringement litigations. Publication without copyright notices by the author, or failure to renew, meant a release of the work into the public domain, without copyright. This burden of compliance put independent creators at a disadvantage relative to companies [Fis15a].

The law offered a provision for composers to assign their copyright to publishers, e.g. in exchange for royalties on sales of copies. An assignment was valid for 28 years, after it could be renewed by the

(15)

composer, or the family if the composer was diseased. If the a song should become popular during the first period, the composer’s negotiation power would have improved by the time of renewal. This created incentives for publishers to encourage the composer to commit to renewing the contract in initial the agreement. Observations of these behaviors triggered discussions about the balance of power between composers and publishers. In the case of Graff, Fred Fisher Music vs. M. Witmark & Sons, Judge Jerome Frank, writes:

In considering those facts, we should take judicial notice of the economic capacities and business acumen of most authors. In ascertaining that certain persons, because “often under economic compulsion,” constituted a “necessitous class,” our present Chief Justice, in 1928, used, judicially, off-the-record knowledge which no casual inquiry would elicit - such as the contents of a report by the City Club of New York, of the Proceedings of the Association of Governmental Labor Officials, and the like. We need not go nearly so far here. We need only take judicial notice of that which every schoolboy knows - that, usually, with few notable exceptions (such as W. Shakespeare and G. B. Shaw), authors are hopelessly inept in business transactions and that lyricists, like the defendant Graff, often sell their songs “for a song.”

Frank means that the law should not permit authors and composers, who are sometimes vulnerable, to sell their rights. The counter-argument to Frank’s position is that authors and composers should be regarded as independently responsible individuals and not treated with special care. Fisher points out that this debate, protection of the authors’ rights against own actions versus individuality, is a recurring theme in copyright law in both Europe and America.

The legal provisions of copyright assignments, however, remain in effect and the term of protection has been repeatedly extended, most recently in 1998¹. In rough terms, copyright assignments make music copyrights into tradeable goods, and enable media companies to buy and sell these rights.

American law provides termination rules, which enable authors to terminate copyright assignments, but these rules are complicated to carry out for artists. Furthermore, artists themselves might not own copyrights to the recordings, see 2.2.

Copyright catalogs Copyright assignments, and consolidation of record companies under major labels, allowed the accumulation of large copyright catalogs. There are currently three major labels;

Sony BMG, Universal Music Group, and Warner Music Group, who together control copyright catalogs that are believed to correspond to 75-90% of streams on popular streaming services [Tsc15;

Ing16].

2.2. Industry actors

This is an overview of typical relations between actors in the traditional music industry [Fis15b].

Composers/artist are writers of original music, but lack the ability to record and distribute. They assign their copyrights to a publishers in return for a royalty agreement.

Publishers then sell recording licenses of the work to record companies. They may also sell other types of licenses; such as for live performances, inclusion in movie soundtracks, or for sub-publishing in foreign countries.

Record companies hire musicians, performers, and produce recordings of the work. They retain the copyright of the recordings.

1The Copyright Term Extension Act (CTEA) makes copyrights valid 50 years after the death of the author, or 75 years for corporate works. This law is sometimes nicknamed “The Mickey Mouse Protection Act”, alluding to the fact it seems carefully formulated to enable an extension of the protection of Disney’s rights to Mickey Mouse.

(16)

Performers and technicians produce recordings of the work. The recordings themselves generate copyrights. It is customary for performers and technicians to have contracts that assign all copyrights to the record company.

Performance rights organizations (PRO:s) acquire performance licenses from publishers. PRO:s then sell blanket licenses to, for example, radio stations and webcasters for broadcast performances. In USA, digital broadcasts of recorded songs require broadcast licenses from the holder of the recording copyright, typically a record company.

Streamers need license agreements from rights holders. This typically means record companies, but may involve publishers and performers.

Digital Millennium Copyright Act (DMCA)

The DMCA is a law passed 1998 in USA and is relevant to us for two reasons 1) the so called Safe Harbor provisions, which limits the liability of service providers from user’s actions, and 2) makes it criminal to bypass copyright protection mechanisms. An equivalent EU directive² exists.

DMCA Safe Harbor “Safe Harbors” are a set of conditions defined in OCILLA³that specifies criteria service providers must meet to limit their liability of their users’ actions. In summary, service providers must implement methods to remove infringing material, block repeating infringing users, and offer technical measures for rights holders to protect their works [VvY12].

2.3. Streaming and downloads

When music is distributed on physical media, consumers are said to “buy music”, although strictly they are only granted a limited license. Similarly, when consumers download songs from online music stores, they “buy” them. Both with physical media, and download services such as iTunes Store, users have the experience of looking at cover art, “buying music”, and “cherishing” their collections. The business model and user experience of downloads was built on the ownership metaphor. Streaming and subscriptions brings a new set of models and metaphors.

Music distribution can be categorized in three models [Wik12]; 1) ownership, 2) access, and 3) context. The access model is exemplified by early p2p networks, and online services such as Apple Music and Spotify. In the context-model of distribution, music is made available to consumers at a certain time and place - e.g. as search results or recommendations. This is exemplified by services such as Spotify.

The Recording Industry Association of America (RIAA) reports sale statistics in categories of distribution. For us, downloads and streaming are the two relevant categories. Downloads and streaming are technically equivalent, but are differentiated in user experience, restrictions, and pricing. A

“streamed” song (legally sometimes referred to as an “ephemeral download”) is technically downloaded but restricted by the playback software.

In the download-model, users experience a type of ownership. There is no limit to where and how many times a download may be played, and users can transfer the songs between devices.

Streaming generates revenue per play rather than per download, which requires that each playback is tallied by the streaming system. To ensure this, playback must only be possible through sanctioned software or devices⁴.

Because licensing grants the streamer rights to stream songs only under specified conditions, the streamer must guarantee confidentiality of the songs. This means in practice that these functions must be vertically integrated, meaning one Streaming Provider needs to operate all of the functions. The

2Directive 2000/31/EC

3Online Copyright Infringement Liability Limitation Act (OCILLA)

4See the “Authorized Distribution” clause in [Sony11, pp. 27–29].

(17)

dominating position of the major media companies has divided the streaming market into a few large vertical partitions, which compete with each other for favorable deals with the media companies.

Streaming and subscriptions There are two common pricing models in streaming; ad-supported and subscriptions. In a subscription model, unlimited access is offered for a recurring fee. The subscription model is enabled in streaming by that fact access can be revoked at the end of the subscription period, which in turn depends on the streaming platform being closed (usually a mobile app).

In practice it has been technically possible, although inconvenient, to obtain files from streaming services that are playable outside of platform specific software.

Streaming royalties License agreements are typically complicated [Sony11], and what follows is a simplified model. Total revenues are shared with rights holders in proportion to the relative number of streams of the song. Typically, the payout for the song X would be given by the formula:

P ayout (X) = # (Streams of X)×T otal streaming revenue

# (T otal streams) × k , 0.75 < k < 1

Where k is a constant determining the share of revenues that the streaming company retains. The precise receiver of the payout, and how it shared between composers, record companies, and publishers is another matter.

The requirement for this model is that stream counts are reliable.

2.3.1. Peer-to-peer networks

Peer-to-peer (p2p) networks enable transfer of data directly between peers without central nodes.

Napster, a famous p2p network, was launched in 1999 and it popularized the idea of sharing music files between hundreds of thousands of people on their personal computers. The Napster service used a central database for keeping record of which files were stored and available for sharing on each of the peers. Users could search this database and download files directly from other peers. The company operating the network and the central servers was targeted by law suits, and ordered to close in 2001.

Gnutella is a p2p protocol for searching and distributing files without a centrally maintained database. This makes the network and service itself an impossible target for lawsuits. BitTorrent, released in 2001, is a p2p protocol designed primarily for resource efficient distribution.

Looking at a chart of the size of the Gnutella network (B.2), we see that the number of leaf nodes peaks in 2011, indicating that casual use starts to wane at that time. The rise of p2p-networks corre- lates with the introduction of home broadband, and starts to wane with smartphones and streaming becoming commonplace. P2p demonstrated the use-case of instant access and user-published content, but is not technically relevant in settings such as mobile networks.

2.3.2. Legal digital music 2.3.2.1. iTunes Store

iTunes is a music player and jukebox software by Apple [iTun; iTuS]. When Apple launched the iPod in 2001, iTunes became a necessary tool for managing music on iPods. iTunes Store launched in 2003, as a complement to both iTunes and iPod, and as a legal alternative to p2p networks. It quickly became one of the most popular online music stores. Which songs were available in iTunes Store was determined by redistribution license agreements between Apple and the media companies. It was not open to independent artists that wished to publish on iTunes Store.

DRM Songs downloaded from iTunes Store were altered by a digital rights scheme (DRM) called FairPlay. A DRM scheme is a technique to prevent media content from being played in unauthorized ways. Incorporating DRM in iTunes was a condition of the media companies’ license agreements.

(18)

FairPlay works by encrypting the audio stream with a symmetric cipher under a key that is included in the file itself, but encrypted under an individual user key.

Because FairPlay is proprietary, only Apple products could play songs downloaded from iTunes.

However, the security of FairPlay was broken within a few years of its release. Examination of FairPlay in iTunes has revealed a surprisingly naive implementation, indicating that it is likely to have been an afterthought, and included mainly to appease the media companies. It was seen as an inconvenience by users.

For a while iTunes Stored offered non-DRM files for a slight premium, but DRM was removed altogether from iTunes Store in 2009.

Apple Music is a streaming service launched 2015, which uses FairPlay for locally cached files.

2.3.2.2. YouTube

YouTube, launched in 2005, is a video sharing site allowing registered users to upload videos for anyone to watch [VvY12]. The business model is in part dependent on the possibility for users to gain income (via ad revenues). Thus, there exists incentives for users to upload protected content.

Interpretation and fulfillment of the DMCA Safe Harbor criterias are central to YouTube, som har drabbats av stämningar för skyddat material som användare har laddat upp.

To meet the DMCA Safe Harbor criteria, YouTube created Content ID, an automatic filtering system that monitors all uploaded content. Technical details of Content ID are not publically available but it can be assumed to function with some form of audio/video fingerprinting. Content ID is open to rights holders that can demonstrate their need and their rights. Approved users are given access to an interface where they can upload reference content that is to be filtered. When the system detects infringement, the rights holder can choose whether to limit showing, mute the audio, or share revenue.

Because of the volume of uploads to YouTube, Content ID, is a necessary system to meet the DMCA Safe Harbor criteria.

From 2007 to 2013 YouTube was engaged in a court case with Viacom (rights holder), where Viacom accused YouTube of deliberately allowing copyright infringements [VvY13]. Viacom supported their argument partially in the fact that Content ID is only effective for rights holders who participate in the scheme and provide reference content. YouTube and Content ID was however deemed to have met Safe Harbor criteria.

2.3.2.3. Grooveshark

The music streaming service Grooveshark was launched in 2007 as a p2p-network, in the style of earlier networks, but in addition to allowing users to share music [Men11], it also offered the ability for users to buy and sell music from each other. The technical solution later departed from p2p, but continued to allow users to buy and sell music. The content on the service was in entirety supplied by the users.

Grooveshark itself received revenues from ads, and a premium subscription service which removed the ads.

The major media companies discovered copyrighted content on the service and a series of law suits ensued. Grooveshark based its defense on the DMCA Safe Harbor, but was not deemed to have met the criteria for monitoring and removing copyrighted content.

The implicit strategy that Grooveshark seems to have pursued [UvE14] (also made explicit in quotes from employees and management [Men11]) was to ignore copyright law and quickly grow a large enough user base for the media companies to fall into a weak negotiation position. This would force them into enter license agreements that would be beneficial to Grooveshark. Therefore, they incentivized users to upload protected but popular material. The service was ordered to close in 2015.

2.3.2.4. Spotify

Spotify, launched in 2007, is a music streaming service with two pricing models; 1) free access with ads, and 2) monthly subscriptions. Spotify, and similar streaming services, do not allow independent creators to upload music directly to their service. The greater part - 75-90% - of all streamed content

(19)

is licensed from the three major media companies [Tsc15; Ing16], each of which Spotify negotiate deals with separately. Independent creators who wish to publish on Spotify are referred to either record companies, or so called “aggregators” whose effective role can be interpreted as ensuring that user uploads do not infringe on protected content, and in a sense shield the streamers from liability.

Technically, Spotify was originally conceived as a p2p network, where users provided a large part of the network and storage capacity [KN10]. Users could play music, which was downloaded and stored in local cache, and music in local cache was made available to p2p peers. Users, however, could not upload or download separate files. (As mobile networks are weak in bandwidth the p2p scheme was not used for smartphones, and is not in use at all anymore.)

To prevent users from exploiting their local caches, music files were “secured” with a DRM scheme which restricted playback to Spotify’s client. The client software is obfuscated to impede reverse engineering.

In general, obfuscation as a security measure breaks Kerckhoff’s principle. Because the pricing model makes it necessary for Spotify to reliably tally the number of plays of an individual song, it must guarantee that no plays occur outside of the Spotify client. Details in a license agreement with Sony obligates Spotify to limit playback to authorised software and devices [Sony11, p. 27]. For a while, a software library was available to developers of third party applications and devices and was usable with premium accounts, but is now discontinued.

2.3.2.5. SoundCloud

SoundCloud, launched in 2008, is a service that allows registered users to upload music, follow other users, leave comments, and other social functions. For a period, there was no way for users to receive income from the service, and therefore small incentives to upload infringing content.

SoundCloud has three main sources of revenue 1) ads in a phone app, 2) ad-free subscription phone app, and 3) paying users who are given unlimited storage for uploads, and detailed statistics on listeners. Revenue is shared with creators in a special program that some creators/rights holders are invited to participate in.

The service has both automatic filtering of user content, and manual mechanisms for rights holders to

“take down” protected content. A large dissatisfaction has been voiced over SoundCloud’s automatic filtering, as it often makes false positive identifications. This is particularly irritating to paying customers.

SoundCloud is pursuing license agreements with major media companies, whose artists then would become part of the service.

2.3.3. Service models and strategies

The preceding cases revolve around two basic service models for streaming. One model centers around user submitted content, and is showcased by SoundCloud and YouTube. The other model is exemplified by Spotify, which is primarily a delivery mechanism for licensed catalogs. In addition, there are two basic use-cases; 1) non-interactive playback mainly on smarthpones, and 2) an interactive experience with social features such as following and liking, not limited to smartphones.

SoundCloud allows users to both upload music and receive income. The number of regular users on SoundCloud exceeds that of Spotify, and the amount of content on SoundCloud vastly exceeds Spotify’s. However, SoundCloud’s revenue-per-user is miniscule in comparison. SoundCloud’s social features are valuable primarily to independent creators and their fans.

Spotify’s revenues mainly results from subscriptions driven by licensed catalogs, and the user experience is streamlined for non-interactive playback.

The subscription model makes all content into public goods. This is not only disadvantageous to artists who wish to expand their fan bases and promote new music, but also to established artists who wish to monetize their current fan bases and song catalogs.

There is a conflict between these two service models, as the major labels and independent creators have opposing goals. The major labels want to use subscription services as streamlined one-way

(20)

output channels, whereas independent creators want to build audiences. The major labels have no incentive to mix their content with that of independent creators, and they can not benefit from social features. Because the catalogs are so important for subscription revenues, major labels can even stipulate conditions that favor their content over other ⁵. The profitability of Spotify depends fully on negotiations with the major labels.

Customer acquisition and subsidies Considering the long term value of a loyal subscription customer, it is profitable to subsidize new customers as an “entry point”. Ad-supported versions of streaming services, possibly with degraded quality, function as entry points. Apple Music offers subscriptions with the first three months subsidized.

2.3.4. Externalities

Playola, a portmanteau of “playlist” and “payola”, is now being used in accusations against streaming services or popular playlist curators, of selling playlist placements [Bil15]. Although streaming services themselves might not be directly selling placements, recommendation algorithms might be configured to infer rankings from certain playlists curated by major label hirelings. Considering that streaming companies are at mercy of media companies for licensing deals, such accusations can not be rejected on implausibility.

A further indication of such practices occurring is the discovery that, for years, Viacom hired marketing agencies to upload copyrighted content to YouTube. To give the appearance of organic fandom, they would degrade video quality and use internet connections untraceable to Viacom [Lev10].

Of note is the “advertising inventory” clause in the 2011 contract between Spotify and Sony Music [Sony11, pp. 36-39] which gives Sony an amount of free ad space. In addition, at least 10% of Spotify’s unsold advertising space would be alotted to Sony content.

These practices recognize that streaming services are not only important for content distribution, but they are platforms for promotion and establishment of new trends. In light of this, there exists incentives for large content owners to have hirelings with disgenuine accounts affect stream statistics.

In fact, since subscriptions are fixed in costs but unlimited in streams, where each stream generates revenue for the content owner, for a certain minimum number of streams, this strategy is self-financing.

Lock-in As with many services, a subscription is not a long-term commitment, and the switching cost for streaming customers is close to zero. Lock-in, as a way of creating switching costs, is therefore an attractive strategy. Examples; making the users feel they are “investing” into provider-specific data such as playlists and preference profiles; linking payments to services that customers already have commitments to.

2.3.5. Asymmetries

The structure of the market results in a deficient equilibrium; one that is suboptimal for the majors, the streaming companies, and the users.

Because license deals with each of the three majors are necessary for any one streaming company to be viable, the majors are guaranteed dominant positions as long as the streaming market grows in total. However, each of the largest streaming companies represent large, but different, portions of the market and important revenue streams. If one streamer were to grow large enough to become dominant, the majors would lose negotiating power. The best strategy for the majors is therefore to demand conditions that prevent any one streamer from growing faster than the market. In fact, this is the dominant strategy for the majors. If a streaming company was to decline a deal with one of the majors, it is likely that the streaming customers would move on to another streaming service. Revenue for the major label would likely not fall by much, but at the very worst by a factor of ¹/#streamers, whereas revenue for the streaming company would likely decline so much as to make it nonviable.

5The “advertising inventory” clause in Sony’s agreement with Spotify [Sony11, pp. 36-39], stipulates an amount of advertising space to be available for Sony’s content.

(21)

Rights Holders

Streaming Companies

Users

Figure 2.1.: Asymmetries in the streaming value chain

In this setting, the best strategy for the majors is to support many competing streamers but restrict any one of them from growing.

A good strategy for streaming companies would be to collude, and negotiate collectively with the majors. Together the streaming companies represent a controlling portion of music industry revenue (see B.1). In practice, collusion among streamers would take the form of an assurance game, which would likely lead to overall defection since streamers are individually in chicken games with the majors - with negative payoffs. In addition, if collusion required individual negotiations between all streamers, transaction costs might exceed the benefits.

Welfare and consumer surplus

These conditions cause inefficiencies. License fees and royalties must be held at the level that restricts streaming companies from growing. Streamers in turn “pass on” this fee to users, or absorb it as losses. Because of the availability of ad-supported streaming and entertainment substitutes, demand of paid streaming is likely to be elastic. Overall market revenue is therefore likely to increase if prices were lower or subscription terms more differentiated.

Streaming companies pursue growth strategies to achieve controlling market positions and even bigger economies of scale. Examples of such strategies are subsidizing customers and financing (fore- seeably unprofitable) growth with debt and new capital. Such commitments can also be interpreted as a tactic in Chicken Games.

Users must choose between with multiple streamers, each providing commodity services, but pursuing lock-in strategies.

Platform economics

Streaming companies are platforms that enable linkages between different classes of customers. A platform needs to satisfy the needs of at least two different classes of customers, and the service provided to each class of customer is in part supplied by other classes. It is critical for a platform that the value proposition to each class of customer offsets the costs of participating. Since value is provided by participants of other classes, platforms need to reach critical mass to be sustainable.

Typically, platform businesses lower the transaction costs for participants who would otherwise not trade.

(22)

However, the need to negotiate blanket license agreements between each streamer and each rights holder results in significant transaction costs.

Frictions and transaction costs

Prevailing streaming platforms introduce several types frictions between listeners and rights holders.

The source of these frictions lies in limitations of the prevailing streaming set up. The terms of the streaming licenses require that streamers control the playback mechanism in order to prevent unauthorized use, which means proprietary and closed technologies. The available strategies for streamers are then lock-in and fast growth to achieve controlling market positions. However, controlling market position for a streamer conflicts with the interests of content owners, and lock-in conflicts with the interests of end users.

2.4. A networked ecosystem for streaming

Imagine instead there existed an ecosystem where the relative size of one participant did not afford it strategic advantage or control over others, and where the collective influence of many participants was not blocked by barriers or choke-points. Both large and small participants would benefit from this set up. Participation would be free and voluntary. The technical platform and specifications, which all participants rely on, would be public goods. This would mean that new participants can invest in the system, trusting that the technical specifications will not change to cripple long-term projects.

Entrepreneurs can build on existing technical solutions, bringing down costs of entry, increasing competition and innovation.

Streaming Ecosystem Majors

Ind epen

dents

Users

Figure 2.2.: A networked ecosystem for streaming

Rights holders would be able to expose their content to the full demand of the market without concern that any one streamer would become dominant.

(23)

Interlinking and transaction costs

If multiple actors in markets with network effects interlink, thereby increasing the effective size of the user set, the value for each individual customer increases. The transaction costs of interlinking must be offset by the increase in the size of the total user set.

The need for individual negotiations can be replaced by a network standard that actors can choose to accept. The internet protocols are examples of this, where adoption of the protocol in effect increases the network size without negotiations. A protocol may be understood as a voluntary pre-negotiated agreement.

Pricing models

The ability to revoke access to streamed songs requires obfuscation and tight control over the technical platform. However, an ecosystem which allows independent actors voluntarily participate in the technical platform can not rely on obfuscation as the security model. Therefore, a subscription based pricing model has to be enabled by some other argument than the ability to revoke access to already streamed songs.

The first argument we make is that revoking is a redundant strategy. The simultaneous growth of streaming, and decline of peer-to-peer filesharing starting from 2011, coinciding with growth of smartphone usage, indicates that convenience is what attracts people to the subscription model. It has been technically possible to obtain non-DRM songs from streaming services, but few users are served by this. Downloads have been shrinking as a revenue segment since 2013 while streaming has been increasing, affirming that it is convenience in the user experience that drives streaming. As long as the intended use of a service is more convenient than subverting it, it seems will be used as intended.

Good incentives are more important than constraints.

Secondly, the subscription pricing model is not optimal in the “pareto” sense. In effect, it makes low- volume users subsidize high-volume users, and it erases the price signal. The defacto per-song price for low-volume listeners is much higher than for high-volume listeners. This could mean that fans of niche content is subsidizing mainstream content, considering that number of streams per song determines royalty payouts. A class of users we might call high-volume/low-commitment, i.e. constantly running background music from mainstream playlists, tally up a disproportionate count of streams for music that is not intentionally selected - skewing payouts to volume playlists.

Price differentiation and discrimination Subscription pricing makes all streaming content into a form of public good. The ability to set individual pricing; for example free access during limited promotion periods would benefit both artists and users. Established artists, with more loyal fans, would benefit from being able to set higher prices.

(24)

3. Towards a new ecosystem

In this chapter, we investigate ways an ecosystem like the one we described thus far can be implemented.

Roles and actors

Rights holders Since copyright law requires that redistribution content must occur with permission of the copyright holder, rights holders need to play a role in the ecosystem. A rights holder must be able to make pricing decisions, distribute files, and receive payments. It is not intended that individual composers or artists fulfill this role independently; rather, they delegate their rights to special service providers, and are free to change service providers at will. Details of this delegation are deferred.

Streaming providers Storage capacity, network bandwidth, and computational resources are required to. Since these scale proportionally to user and geographic demand, independently of rights holders, these requirements should be fulfilled by a separate role. The streaming provider contributes resources, and must receive payments in return.

Payment providers While the system keeps records of transactions, payment providers are required to settle payments between users, rights holders, and streaming providers with money. A payment provider takes money from end users in return for deposit certificates, which users exchange for services.

Providers of these services can use transaction transcripts to claim money from providers.

App and device makers The user experience of the service will be provided by app makers. An app is a commitment to both the ecosystem and the specific technology target, for example a mobile phone platform. Some target platforms such as mobile phones allow frequent updates, which makes it possible for app makers to adapt their applications to an evolving network service. There are embedded platforms, e.g. car sound systems, clock radios, or stereo units, that don’t permit software updates.

In these cases the viability of the application depends on the long term stability and predictability of the network service.

For app makers to invest effort into the system; a sizeable user base, a stable and predictable technical platform are all required.

Users End users listen to music from their diverse and distributed devices. They might be limited network capacity or bandwidth.

Other roles Through these technically defined roles, various business models may be articulated. For example; an ad-brokerage might provide users with deposits (like payment providers) in exchange for listening to advertisements. Similarly, various pricing models and promotion schemes may possible.

Market actors such as streamers may operate all these roles and present a single branded experience to their users.

Critical mass and cost of participation

Platforms, where one class of user provides services to other classes, have network effects. Since value of participation depends on the number of participants, if there is a cost of participation, there is also a critical mass that must attained before the network is self-sufficient. This cost is usually covered through subsidies until critical mass.

A Buyer-Seller Protocol with Watermarking for Cloud Streaming

A Buyer-Seller Protocol with Watermarking for Cloud Streaming

Johan Björklund

June 2017

Contents

1. Introduction

Purpose

1.1. Research question

1.2. Methodology and scope

1.3. Related work

1.4. Contribution

1.5. Structure of this report

2. Music distribution

2.1. The economics of music

2.2. Industry actors

2.3. Streaming and downloads

2.4. A networked ecosystem for streaming

3. Towards a new ecosystem