Navigation – Plan du site
Dossier thématique

Towards a micropolitics of formats

Content id and the management of sound
Vers une micropolitique des formats. Content ID et l’administration du sonore
Hacia una micropolítica de los formatos. Content id y administración del sonido
Guillaume Heuguet
Traduction de Fanny Quément
Cet article est une traduction de :
Vers une micropolitique des formats [fr]


Cet article croise des perspectives issues des études de sciences, des sciences de l’information et de la communication et des cultural studies. Il s’intéresse au dispositif Content ID édité par l’entreprise et site de publication de vidéos YouTube, qui permet d’identifier et de monétiser les morceaux de musique présents dans les vidéos. Il s’agit d’analyser la construction d’une innovation dans les marchés de la musique et de l’Internet. Ce dispositif s’inscrit dans la filiation d’enjeux stratégiques et de techniques liés à l’industrie musicale et à l’économie du Web, qui concernent l’arbitrage entre circulation des formes culturelles et leur contrôle comme œuvres et marchandises par le droit d’auteur : dans ce cadre, Content ID est présenté comme un compromis idéal. La mise en œuvre de cet idéal repose sur l’ancrage de l’œuvre musicale dans le son, sa modélisation informatique, et la naturalisation de ce modèle. Au-delà d’un équipement technique des règles de marché, il s’agit d’une semi-automatisation de jugements culturels et juridiques d’originalité. Content ID redéfinit ainsi discrètement les valeurs de la musique enregistrée, et la façon dont elles sont instituées.

Haut de page

Entrées d’index

Palabras claves:

Internet, informatica, formato, YouTube
Haut de page

Texte intégral


1Science and technology studies have long drawn attention to the implicit "moral" dimensions of technical devices. They have highlighted their ability to organize social consensus by defining possible situations and interactions (Latour, 1996). They have also helped realize how much routine can mask the agency of technical devices, and how the human decisions that governed their configuration then became "black boxes" (Latour, 2005). These approaches have inspired a variety of works in fields such as market instruments and devices (Callon et al., 2013) or image and sound compression techniques (Mackenzie, 2006). As for French research in information and communication sciences, it has been focusing on the semiotic and media dimension of computer and web technologies, i.e. how they combine different writing standards from computer programming to the organization of signs on the screen that prefigure and condition practices (Jeanneret, 2007). Therefore, they invite to add to the description of socio-technical arrangements the interpretation of the forms through which these arrangements are instituted and interpreted (Jeanneret and Souchier, 2002). It could be argued that the common point of these perspectives lies in a concern about the material, technical and media conditions of economic and social exchanges, with a particular interest in the ability of entrepreneurs and engineers to impose their visions of the ideal social organization, taking into account the opacity of the techniques and the thickness of the mediatization processes (Jeanneret 2001). These issues are also central to the work of Jonathan Sterne, a champion of cultural studies whose approach blends political economy with industrial history and post-Foucaldian perspectives.

2These perspectives make room for interest in the micro-political dimension of technical and media formats. For Jonathan Sterne, the study of formats "is based as much on small mechanisms software, operating standards and codes as on infrastructures, consortia of international companies or entire technical systems" (Sterne, 2018). A micro-political approach to formats thus makes it possible to bridge the gap between approaches that are socio-technical and/or communicational, and part of cultural studies. It connects the investigation of economic innovation strategies with the study of design, and more generally with the part of imagination and poetics that comes into play in the creation of cultural techniques and forms.

3This article falls within this framework. Its focus lies in a media apparatus which has become hegemonic, particularly in the field of music: YouTube. Launched in 2005 by three former PayPal online payment employees, Alphabet Inc.'s subsidiary publishes a site and application primarily designed for hosting and publishing videos, which became the first online music listening destination for the French according to the Hadopi (Higher Autority For The Dissemination of Works and Protection of Rights on the Internet)1. With a partnership allowing it to distribute catalogs from three of the biggest majors through Vevo, a labeled video system, YouTube hosts a large database of clips by successful artists and makes it available at any time (McIntosh, 2016). It completes a no less extensive catalog of songs and videos published by Internet users or other companies, which differentiates YouTube from other streaming software such as Spotify or Deezer, whose catalogs depend on the gatekeeping of major distributors specialized in digital music. On YouTube, the management of the commercial and legal rules regarding this catalog depends on a technical system called Content ID, which subsequently detects the presence of sounds and allows right holders to order their removal or take advantage of the audience they attract and YouTube's advertising network to generate revenue.

  • 2 For a genealogy of this construction in France, cf. Beuscart (2006).

4It is this system that I intend to analyze more precisely, at the crossroads between two issues that link the sociology of science and music knowledge: the construction of an innovation in the music and Internet markets2 and the redefinition of the category of the work in the terms of a project to identify and control musical pieces.

  • 3 The following analysis uses english-langage documentation regarding US-based policies, while source (...)

5To describe this dual process, one must read the strategic and promotional speeches published by the YouTube team in the light of the history of online music, study the patents and technical documentation about the systems implemented, and focus on the legal texts, legal relationships and judgment reports that define the criteria to assess the originality of works and the purposes for which their circulation is controlled3.

6I will first show that YouTube’s development of Content ID is at the crossroads of strategic issues and techniques related to the music industry and the Web economy, more specifically the trade-off between the circulation of cultural forms and their control as works and goods: Content ID is a technological compromise for the management of copyright and related rights online. It will then become obvious that part of the system is based on anchoring the musical form in sound, through a modeling approach to music, sound and listening that leads to the "naturalization" of the work in computer data. Finally, I will describe how YouTube sets up a mathematics of originality, which has consequences not only in regulating the circulation of works subject to copyright, but also regarding the definition of public music culture and its values.

From the control / dissemination dilemma to a technological compromise

7The commodity status of digital music has been evolving since the late 1990s, through milestones such as peer-to-peer exchange (Napster was launched in 1999), the rise of music blogs and free hosting and file transfer platforms such as MegaUpload (around 2005), the way majors experimented with hardware copy blocking (DRM, between 2001 and 2006), the success in selling music files from Apple's iTunes Store with its fixed "€0.99" price in parallel with the launch of the iPod, the development of competing local platforms (in France,,, the introduction of Hadopi sanctions (2009), or the development of on-demand listening players: Spotify was launched in Sweden in 2008, Deezer obtained rights holder agreements in 2009, and in the same period, YouTube started being used to listen to pre-existing directories.

8In this process, the digitization of music has been reinforcing the status of recorded music as an emblem of "transsectorial commodity" (Théberge, 1997). Indeed, whether it circulates as MP3 files via peer-to-peer software or as "data packets" erased as they are played via streaming sites and software, music is taken into multiple networks of transactions and equivalence. A piece of music can be sold as a file or function as a commodity "by destination" (Appadurai, 1986) and be included in a catalog in order to sell a subscription to a site or listening software; be offered free of charge to generate "clicks" or "views" and thus build a capital of notoriety or visibility; or simply serve as a "premium product" to attract Internet users and encourage them to register or contribute (Bouquillion and Matthews, 2010). Internet users are then counted to become "commodity audiences" (Smythe, 1981) and accordingly, media devices based on publication and networking tools (Jeanneret, 2014) blend the advertising economy of radio and television with the speculative financial model of "platform capitalism" (Srnicek, 2017).

  • 4 The "theatrophone" was supposed to allow, over a century before the streaming boom, listening to op (...)

9Online music therefore brings together at least two extensive networks of music circulation in which its social and economic value lies (one way or another). The first of these networks is a music market, whichever form it takes: the authors or the publishers and users of the works are the ones who are ultimately interested in their circulation or control. The second has to do with what could be called the innovation market: all technological or media products and infrastructures that update their potential through music, using it or allowing it to be used. . At least since Thomas Edison, recorded music has served to develop a market for devices and networks, some of which have been very successful to date (such as the phonograph, the radio transistor, cable television networks) while others have been neglected by most music and technology scholars (like the jukebox) or simply never made it, such as the forgotten "theâtrophone", which was supposed to offer special music programs4.

10Similarly, YouTube managed to benefit, for a while, from the attractiveness of music and thus drew an audience around its video hosting project, which gradually became a media in its own right with a wide range of initiatives and micro-devices developed partly in connection with music: ceremonies like the YouTube Music Awards, audience analysis tools such as YouTube for Artists, mood playlists (Moods), recommendation systems first promoted in reference to musical discovery, crossed with continuous playback functions (YouTube Disco, now YouTube Mix) (Heuguet, 2018). To this end, the company quickly reached agreements with the main record majors interested in Google's resources, which gave it a head start over its competitors in this field (Mulligan, 2015). It could also rely on its community rhetoric to appeal to independent labels and musicians, who saw it as an original means of promotion, with a ripple effect for the most famous record companies and artists: each year from 2007 onward, YouTube’s most viewed videos were generally from musicians signed by one of the three majors, such as Shakira, Avril Lavigne or My Chemical Romance5.

  • 6 Introduced by law n ° 85-660 of July 3, 1985 in France, neighboring rights extend rights of authori (...)
  • 7 Law n° 2004-575 of June 21, 2004 « Pour la confiance dans l’économie numérique ». https://www.legif (...)

11The de facto agreement on the free distribution of controlled music was short-lived. Regarding national and international legislation, although several experts claim the service includes editorial dimensions (European Audiovisual Observatory, 2007), YouTube can claim to be a host rather than a publisher, which allows the company to remove productions that are freely circulating on its servers and pages while being subject to copyright and neighboring rights6, only when expressly requested by the right holders: in the United States, this is known as the “safe harbor” system and its “Stay down & notice” logic (Heald, 2014), adapted and enshrined in the 2004 French Law for Confidence in the Digital Economy7.

12Although the music industry and the majors did benefit from YouTube as a publication and promotion space, their best interest is to ensure that any use of their catalogs becomes a source of profit. In that regard, tradition is on their side. Record publishers, after a period of adjustment, had already succeeded in imposing the re-qualification of the radio-broadcasting of musical works as a full exploitation rather than a mere opportunity for promotion. However, experts suggest that the Internet and the Web have a specific potential requiring reflection on how to monitor and assess them. For example, in a book entitled Copyright's Highway: From Gutenberg to the Celestial Jukebox, Stanford law professor Paul Goldstein speaks of the "celestial jukebox" (Goldstein, 2003), a phrase linking the charm of the old device with the utopias of the intangible (Gomez Mejia, 2014), to convince media conglomerates "that with some adjustments to copyright laws, they could monetize their catalogue holdings in a new medium full of promise"8. Patrick Burkart and Tom McCourt then suggested to complete customer relationship personalization technologies (CRM for Customer Relationship Management) with copy locking devices (DRM for Digital Rights Management) in order to provide a "new infrastructure" for online music, which is neither peer-to-peer software nor "brick and mortar" stores (Burkart and McCourt, 2004).

  • 9 « To clear up confusion, this is not a new feature. The dispute process has been in place since Con (...)

13Where YouTube has an interest in hosting and circulating as much music as possible via its media player, record companies are therefore looking for new ways to regain control over their catalogs. YouTube's Content ID system emerged as a compromise. After Google’s 2006 takeover of the company, it is "under pressure from [other] companies" that YouTube introduced, in October 2007, a system identifying and tracking registered videos (Kim, 2012). That year, the company tested its extended version under the name Content ID, in partnership with Time Warner and Disney9, but it was with a focus on music that YouTube made it official, by the end of 2008, in a blog post entitled "The Up & Downs of Music Licensing on YouTube". In this announcement, the company formalizes the sound and video track identification system by explicitly presenting it as a way to build a partnership with the music industry:

We work with the music industry worldwide – with major and independent labels and publishers, rights collecting societies, and with artists and songwriters directly – to build user-friendly licensing arrangements. With these partners and our cutting-edge Content ID technology, we’ve created a win-win situation the likes of which never existed before YouTube: a system that gives artists a brand new revenue source and a great way to connect with their fans.10

14However, Content ID did not immediately function as a "win-win" agreement: Warner Music started using it to block the use of music from its catalog, then left the agreement with YouTube in the first year because profits were deemed insufficient (Kim, 2012). In 2009, YouTube filed its own patent for "blocking unlicensed audio content in a video hosting site" and then began to communicate regularly about improvements to the system. A more comprehensive definition of the device was finally given to the general public in April 2012:

  • 11 YouTube, On line: (Re (...)

Content ID allows rights holders to automatically detect uploaded content that contains potentially infringing works. This is possible through a system that creates a digital “fingerprint” of a video when it’s uploaded to YouTube. These fingerprints are automatically compared to partner-provided references. In case of a match, the rights holder can choose to monetize it, block it from the site or keep track of viewing metrics. Because of the vast scale of YouTube ( remember: 1 hour of content is uploaded to the site every second), this is an automated process.11

  • 12 The European directive on digital commerce (applied in France), designed to promote the competitive (...)

15It was then clear that Content ID was a technical response to the recommendations of the Notice & Stay down logic, in the form of a technical system helping YouTube to process all the withdrawal requests from rights holders by automating the procedure. Since the general trend is a progressive strengthening of copyright (Vaidhyanathan, 2003)12, a YouTube product manager found it appropriate to celebrate his company's action:

Pourquoi est-ce que j’adore ce que YouTube a créé ? Parce qu’il permet à des dizaines de millions (des centaines de millions ?) de vidéos d’être améliorées en combinant différents morceaux de contenus. Ensuite parce que c’est une manière élégante de reconnaître que le contenu a une valeur commerciale et si vous utilisez celui de quelqu’un d’autre, ils méritent une contrepartie. Et parce qu’il crée une nouvelle base subversive – une génération grandit en partant du principe qu’elle a le droit de remixer des choses.13

16At this point, YouTube claimed to have succeeded in opening up expression regimes where actions to defend global licenses had failed (Bouquillion, Miège and Morizet, 2006):

Les plus grands changements dans le copyright ne viennent pas du système légal américain mais des entrepreneurs, des technologues et des capitalistes qui voient une opportunité et une façon de faire croître leur marché plutôt que de se battre pour un même gâteau.14

  • 15 Founded by law professor Lawrence Lessig, Creative Commons is an American "non-profit" organization (...)

17YouTube thus discreetly and then triumphantly presented itself as the inventor of a miracle solution to the ideological and legal dilemma of the relationship between the circulation of works and their control. Of course, one will immediately spot potential tensions between the project to make the uses and practices of appropriation an exercise kept under the control of a private company, and the visions of the commons or the public domain that are championed by projects like Wikipedia or Creative Commons15. Making such practices dependent on a private company changes the meaning of the appropriations at stake (Bullich, 2015). YouTube's philosophy, well explained and summarized by Hunter Walk through the slogan "Copy, Remix, Profit", also confuses artistic property with what the French legal tradition has separated as moral rights allowing to prohibit or authorize appropriations independently from any question of remuneration. It also implies that any appropriation of a previous work is, de facto and at any time, likely to take on a promotional and commercial value.

18This conception of market instruments as optimal tools to organize the sharing and promotion of culture has been at the core of the discourse about Content ID, whether it be produced by law scholars (Heald, 2014) or by associations interested in “free culture” issues16. However, it is possible to consider it from another angle, which concerns less cultural exchanges as a whole than the aesthetic and technical imaginaries supporting it, and the concrete means the company provides itself with in this perspective. Indeed, far from representing a simple technical solution offering a consensual and balanced definition of the exchange of music videos as works and commodities, Content ID is part of a genealogy of digital music control techniques. We will see now that it implies a specific ontology of sound and a mathematical definition of originality.

Naturalized sound marking

19When Hunter Walk celebrates "what YouTube has done", he omits the company did not imagine the Content ID system on its own. Even though it initially took the form of video identification (Video ID), YouTube’s first attempt at content identification started with another company, Audible Magic, with which it signed a licensing agreement in 2006, shortly before its competitor MySpace ("Audible Magic Accuses YouTube of Fraud Over Content ID Trademark", 2017). As its name suggests, Audible Magic worked on and with sound recognition tools. Its website now presents it as a solution for "automatic content recognition" and the "de facto standard for music identification for compliance / rights with content recorded by majors and major music aggregators," displaying as reference customers Dailymotion, Facebook, Soundcloud, Twitch or Vimeo17.

  • 18 The circulation of decoy MP3 files, where an artist name and a title in the file name covered an in (...)
  • 19 Blum, T. L. S. F., Keislar, D. F. B., Wheaton, J. A. F. et Wold, E. H. E. C., Google Patents. On li (...)

20In the context of tensions surrounding the free circulation of copyrighted musical works as MP3 files, Audible Magic has developed and filed a series of patents such as “System for distributing decoy content in a peer to peer network”18, “Method And Apparatus For Identifying Media Content Presented On A Media Playing Device", "Copyright Detection And A Protection System And Method", "Method And Device For Creating Unique Audio Signatures"19, “The use of fingerprints to associate data with a work", “Method and apparatus for identifying an unknown work".

21The very name "Content ID" became the object of a dispute between Audible Magic and YouTube, the former accusing the latter of having appropriated a registered trademark. Yet the name originated in a common resource: it is indeed part of the specifications of HTML standards, and in this context it refers to identifiers that allow the use of remote resources, such as "media files" on a web page or in an e-mail (the source code20 of YouTube players displayed this type of code for a while: "id=xxx"). Companies like the antivirus giant McAfee or the microprocessor developers Intel also patented techniques for content identification that have been implemented in online messaging applications such as Yahoo Mail, but they clearly had not considered making their processes registered trademarks21.

22The technical conception of identifiers can also be found in the music world, but without reference to "IDs". It is based on a marking principle, as implemented with the watermark technique to track the ripping and duplication of sound files from discs22. The term “watermark” refers to the use of invisible ink on paper as an image for the writing of data into the data of a piece of music in binary format, in the form of signals that are inaudible but detectable in a digital spectrum analysis. Not only are these data imperceptible to the listener, but they are also designed so that it is difficult to isolate them without affecting the sound23. The watermark thus makes it possible to imprint a mark directly onto the sound signal, so that the circulation of unauthorized copies can be traced beyond format alterations.

23When Audible Magic refers to its technology as a "sound signature", the company uses a term close to the principle of the "watermark": the signature is a deliberate gesture designed to trace and authenticate a source (the signatory), so the authentication operation and its artificial character prevail. Appropriating the technology provided by Audible Magic, YouTube preferred the phrase “digital fingerprint” to “sound signature”. They have a different way of qualifying the relationship between the signal and the work to which it refers. Indeed, the metaphor of the fingerprint assumes an indexical and metonymic relationship, such as the one theorized by Pierce (1991): considering a sign as an index implies there has been a physical continuity between the signal and its source. In this case, the signal is presented as part of the physical phenomenon of the musical work in question, what could be called its “sound matter”: speaking of the "digital fingerprint" of a piece of music would refer to some "sound bite" in the form of computer code. Thus, the metaphor of the "fingerprint" suggests that part of the object itself becomes the sign. This relationship of metonymic contiguity functions in opposition to the operation of "coding" implied by the term "digital", a term which originally refers to fingers before becoming a pervasive metaphor for computer code and processes. Thus, YouTube's "digital fingerprint" is a pleonasm – due to the forgotten roots of “digital” – as well as an oxymoron – there is no “indexical” sign in the mediation of symbols by computers (in fact, the very principle of the Turing machine and computation is based on the abstraction of sign from any concrete and singular object).

24The "digital fingerprint" also contrasts with "watermark" and its fiduciary cultural associations (i.e banknotes), alluding instead to the human body and administrative identification practices. Here, the retrospective operation involved by the watermark – a deliberate effort to mark and singularize pre-existing supports in order to maintain their exchange value or control their circulation – is replaced with a genetic perspective: musical creation is assimilated to filiation (with a reference to the "paternity" of the works). Speaking of a “digital fingerprint” therefore aims at bringing home that a work has a singularity independent from the media supporting it. Incidentally, the idea of a singularity conceived as intrinsic justifies a data chunk exchange value. In the end, the residual common point between this “digital fingerprint” and watermark only lies in the procedure of proof: the fingerprint, even more than the watermark, brings into play the notion that clues cannot be forged.

25This naturalization of sound tagging was later confirmed when, in 2019, YouTube stopped referring to the "creation of identifiers", merely mentioning instead its reliance on a "file database"24. The company therefore gradually withdrew the intentional and artificial dimensions of fingerprint production as a material and symbolic reality (a computer artifact), an issue that was still key to any public discussion on watermarks – especially the various strategies to circumvent them.

26The metaphor of the "digital fingerprint" seems to suggest that the digital code, far from functioning as a model, could eventually be as singular as the organic fingerprint. This way, it obscures the transformations involved in signal processing and the construction of data as "footprints" or traces (Jeanneret, 2011). Such a conception contrasts with the way human and social sciences have been highlighting the mediations and transformations at work in any form of writing, especially the many variants of computer code, if only because this "technical" writing implies its own cultural imaginaries and metaphors. We must therefore distance ourselves from the naturalizing perspective of the "fingerprint" and insist on its oxymoronic aspect in order to better describe what it encompasses: a change in the philosophy and the technical strategies to control works. Indeed, the logic of "marking" has given way to the production of an identifier from the reformatting of sound as a mathematical object.

The reformatting of the sound object

27The detail of the technical system on which Content ID is based is not accessible to us: YouTube Help pages explain the uses and "good practices" regarding the identification tool, without saying anything about its internal functioning25. Nevertheless, we can approach this functioning through techno-scientific research. In the computer world, the tradition of research on the coding of cultural and sensitive forms, whether graphic or sonic is associated with the phrase perceptual hash. "Hash" refers to a conventional symbol used in marking: in practice, a series of numbers (also known as a "key") functions as a unique "identifier". With its reference to the “hash" symbol, marking reveals itself as numbering and the artificial dimension of the operation is made clear. The term perceptual, on the other hand, draws attention to a dimension which remained implicit in previous metaphors for the production of identifiers: the fact that a cultural form only exists in relation to a perceptive activity, which itself depends on thresholds of sensitivity. The problem of the relationship between the "data of perception" and the identification/construction of motifs or forms has been recognized and integrated into the general questions of aesthetic philosophy (Morizot, 2015) and the semiotics of the sensitive realm (Fontanille, 1995); it has been more precisely studied as the relationship between categories for sound and music in the philosophy of music (Charrak, 2001; Kaltenecker, 2010) and the semiotics of listening (Valle, 2015). Despite their diversity, these disciplines and authors generally agree on identifying the problem of the historical and contextual relativity of the thresholds of perception which define the relevant forms from which aesthetic objects and their interpretation are developed, raising a question that is a classic in the history of arts and techniques (Francastel, 1956). However, as Pierre-Carl Langlais points out, the scientific literature on perceptual hash proposes to ignore this historical and contextual relativity in order to formalize sound and model perception26. These studies take over from the scientists who have worked on sound frequency by separating the idea of sound from its multiple fields of application (speech, music) (Kittler, Vargoz, Alloa and Guez, 2018), the engineers who, to invent recorded music, had tried to "mechanize" the hearing process (Sterne and Boidy, 2015) and finally their successors who suggested new psychoacoustic models of the same activity in order to develop compression formats such as MP3 (Sterne, 2018).

28The engineering of perception underlying the creation of the "identifiers" of cultural forms thus embodies the bias that it would be possible to isolate once and for all the contours of aesthetic objects and their sensitive substrates; or rather, aware of the difficulty of establishing once and for all a model of sound hearing and musical listening, engineers and data scientists strive to find a sufficiently satisfactory technical solution to such a challenge.

  • 27 For an exception concerning the genealogical approach to machine vision, and whose results on the " (...)

29As Maria Eriksson (Eriksson, forthcoming) notes, this type of work is mainly dealt with in scientific articles specializing in the fields of "machine listening" (Gras, 2014), "music information retrieval" (music information retrieval) and their main technical challenge, namely "feature extraction". In academic databases, the works classified under these categories are often richer in equations than in reflections on what is at stake in such a project27.

30However, a fairly comprehensive overview of these questions and the methods used was provided by three researchers from the Barcelona Media Technology Research Group (Alías, Socoró and Sevillano, 2016). These authors are not particularly interested in the "fingerprinting audio" project, but in the technique that makes such a project possible: the extraction of features, in the sense of mathematical units representing "information" from a sound signal. The authors point out that the research on modeling musical frequencies is only a sub-section of the set of models that differ depending on whether they are dealing with audible or non-audible sounds; within audible sounds, speech, music or environmental sounds; and within environmental sounds, noises, "natural sounds" or "artificial sounds". Such partition already shows the dependency of the models on a repertoire of norms combining "physical nature", psychology of perception and cultural classification (such as the distinction between speech and music). These operations of differentiation look specifically at the way the signals associated with each category are distributed over the entire audio frequency spectrum as it is most often modeled (from 0 to 5000 hertz).

  • 28 The cepstrum is the transformation of a mathematical modeling of temporality into another mathemati (...)

31According to the authors, the whole range of extraction initiatives also falls into two main groups: those relying on a physicist model of sound, and those working with a perceptual model (as with the perceptual hash). The extraction of musical features is mainly identified with the category of "notes", themselves reduced to the following temporal and spectral characteristics: "from short to long durations (40-1200ms), with a mixture of constant and ephemeral sounds organized into periodic structures, a largely harmonic content throughout the band from 20 Hz to 20 KHz, including inharmonic parts". Finally, the techniques might be divided according to the variables they favor: time, frequency, wavelets, source image or cepstrum28.

32This list of variables could be longer. Some, like the “notes”, clearly include incidental cultural judgments like the ones revealed in the debate on the "objective" existence of tonality and the cultural variations in its recognition. Others are more readily linked to mathematical and computer operations made feasible, such as the measurement of the "pulse clarity" or "energy" of frequencies distributed at different intervals in the spectrum (subband energy ratio).

33Among this wide variety of criteria, we don’t know exactly which model YouTube has chosen. Did the site engineers rely on Audible Magic's method, based on the combination of volume, height, gloss, and "cesptral" coefficient properties, cited in a registered patent (Blum et al., 2008)? Or did they use the patents of the parent company Google, such as this "system and methods for recognizing sound and music signals in high noise and distortion" filed by Avery Li-Chun Wang29, who has since become the founder of Shazam, the most famous application for recognizing broadcast songs?

34Relying on Wang's publications and his own knowledge, the French engineer Christophe Kalenzaga sought to understand the model behind Shazam30. He believes Wang went for "image-based" parameters: in this case, the computer code produced is first of all a mathematical approximation of the coordinates associated with its spectrogram, i.e. a graphical representation of the frequency. The spectrogram offers a partial view of certain musical parameters such as tonality, harmony, rhythms or timbres. These different units of meaning, which structure music theory, practice and composition, are partly translated and partly ignored by the "diagrammatic" representation of the spectrogram. With spectrogram modeling, the final identifier (the "hash") results from the mathematical reformatting and computer recording of a graphic representation of a psycho-acoustic model of sound.

35While there is no certain indication that this solution has been favored by YouTube, it is sufficient as a starting point to pursue our argument and understand how, far from being satisfied with an "objective" model of sound, the work of perceptual hashing constitutes a reformatting. This rewriting is first and foremost a material phenomenon: although the objective is to base modeling on the human perception of sound (to achieve a sufficient degree of "similarity"), the data processing necessarily affects the realization of this ideal, and produces a new object, a specific calculable and writable form. But this reformatting is also a symbolic phenomenon, since in passing, the chosen model embeds a definition of what is relevant in the perception of music. What the reference to the imprint (the fingerprint) presented as the singular, faithful and unfalsifiable indexation of a piece of music therefore corresponds to a relatively dense reformatting operation, rich in scientific, technical and scriptural mediations.

36In scientific texts, the "hash" is either represented as a diagram made of full and empty areas in a historiogram, or a "key" (a series of characters in finite number which serves as "name own "to this first numeric code).

Figure 1. Examples of two "hashes" (for an original image and its thumbnail) represented as a digital key as proposed by the engineer Jens Segers31

Figure 1. Examples of two "hashes" (for an original image and its thumbnail) represented as a digital key as proposed by the engineer Jens Segers31

Segers, J.,

37If we want to get an idea of the density of mediations which contribute to forge such an object, we can also turn to the "perceptual hashes" of images. Indeed, their different states are often represented by “metamorphosed” images: we see their pixelated, cropped, zoomed copies, the perspectives are reduced when they are not the object of an anamorphosis, the contrasts increase and the number of colors decreases…

  • 32 On the criticism of the "averaging" of hearing by laboratory tests, cf. Tern (2018).

38As sound hashing works on the basis of the processing of a graphic icon - the spectrogram - it involves two distinct semiotic systems. There is therefore no access to these "intermediate states" in a sound form, which would allow these alterations to be felt in a homothetic manner. Passing through the visual equivalent of the processing nevertheless makes it possible to consider that hashing does not consist of a reduced model or a transcoding, but that it includes dimensions of filtering, compression and alteration of the model of the perceived sound form - which necessarily involves using an abstract model of this perception, possibly the definition of an average hearing capacity32.

Figure 3a. Two figurations of the mediations involved in the process of producing a "perceptual hash" - figuration by Cano et al.

Figure 3a. Two figurations of the mediations involved in the process of producing a "perceptual hash" - figuration by Cano et al.

Source: Cano, Batlle, Kalker & Haitsma, 2005

Figure 3b. Figuration de Jovanovic33

Figure 3b. Figuration de Jovanovic33

Source: Jovanovic, 2014

39Another way of trying to imagine the processing performed to achieve the "digital fingerprint" corresponds to another type of diagrams, namely the flowcharts well known in the world of engineering patents. The reconstruction of the multiple operations involved exceeds the scope of this article, but these diagrams are sufficient to suggest the extent of the mediations and the “destructive” and “reconstructive” part implied by the operations which exist under the reference to “the digital fingerprint. " We find, in another form, some of the variables identified above, and still others: transition from stereo to mono, filtering of certain frequencies deemed non-determining, application of algorithms, modeling of audible frequencies, selection of certain representative zones in the spectrum, selection of a sampling frequency, extraction of properties deemed to be decisive (note pitch, low frequencies, intensity).

40A peek at perceptual hashing diagrams and flowcharts in patents is enough to get the extent of the mediation and the "destructive" and "reconstructive" phases in the operations masked by the reference to "the digital fingerprint": some of the variables identified above will be found, as well as others like the transition from stereo to mono, the filtering of certain frequencies considered as non-determinant, the application of algorithms, frequency modeling, the selection of some representative "areas" in the spectrum or that of a sampling frequency, the extraction of properties considered as decisive (pitch, low frequencies, intensity). As these diagrams suggest, whichever operation is focused on, it appears the choice of technical means to model music aims at maximum efficiency, but the definition of this efficiency remains a matter of qualitative judgments and involves a multitude of decision-making levels. For example, developers of sound hashing techniques must define a threshold and develop techniques to draw relevant information from the acoustic continuum or, in other words, to separate "noise" from "musical sounds". They must also define alteration thresholds so that the compression or filtering of the sound files, which distort spectrograms, do not prevent their programs from recognizing the "same" song: this sometimes happens when converting from one sound file format to another, as well as with remasterings, edits or remixes, which constitute a neglected but important part of popular music heritage (Everett, 2010).

41There is a variety of algorithmic "solutions" to the challenges posed by the synthetic modeling of a perceived cultural form. In the most common models, which aim at saving storage space and limiting the number of computational operations, a short excerpt from a piece of music is supposed to be enough to identify the whole piece. For the same reason, a limited series of "data points", i. e. digital landmarks in the spectrogram of this “short excerpt”, is supposed to be sufficient to define the model of a singular piece of music. Engineers and data scientists must arbitrate between several definitions of what constitutes effective modeling, in particular between the search for precision (or "granularity") of the model and the search for a minimum of time and therefore computational resources for the calculations necessary to match an "identifier"-source with its target in a database.

42These techniques also face inherent challenges that make them mediations in their own right: for example, the mathematical functions used to reduce the size of the “hash” produced from the spectrogram tend to produce "artefacts" (spectral leakage) that must be countered.

The limits of the work’s anchoring in sound

  • 34 Interview with the head of YouTube rights management for one of the main French digital music distr (...)

43Despite the naturalization of the technical process in YouTube's speeches, the "perceptual hashing" that underlies Content ID operates a computer reformatting of the sound object. We have seen how it involves a series of technical mediations, cultural criteria and strategic decisions. We must now take an interest in the uses of this identifier, which consist in comparing it with references present in the databases. From an operational point of view, this comparison is based on a relevant duration and a similarity score. However, these criteria which show a bias and a trust in probabilities within the matching process exist only for expert teams working to "track" references, particularly among digital music distributors with whom YouTube has concluded agreements34.

44For a wider audience, the metaphor of the "fingerprint" implies that matching is not a new calculation whose factors and margins of error must be re-weighted, but a simple "recognition" of the presence of a musical work or musical fragment in a video. Thus, at this stage, the question is no longer about the precision or robustness of a computer model, or the fidelity of the hash as a sound model, but about associating to this computer model an authentication function for a musical work, which is not quite the same thing: where sound generally refers to an objective physical reality (although we have seen that its singularizing depends on a psycho-acoustic model), musical works are more readily associated with cultural and intersubjective values of originality and authority.

45These values are part of a historical construction owing as much to musical aesthetics as to the legal framework of copyright and related rights. As it relies above all on the score i.e. on the criteria of harmonic writing, arrangement and lyrics the judgment of singularity seems to begin to rely on sonic parameters when it becomes possible to deposit recordings directly with the authors' societies. However, in this context, the musical recording, i.e. the disc, still works like a "testimony": the sound is not confused with the work, because from a material and legal point of view, the phonogram remains a fixation support that re-produces it. It then serves as a reference or witness when it comes to enforcing rights, such as in cases of dissemination, unauthorized public reproduction of part of it, or even sampling or borrowing.

46When the digitized form of a sound recording is the perceptual hash/identifier, the musical work is folded into its mathematical definition. Regarding the detection of "similarities", the physical, tangible record as a technical mediation of the work disappears, and we are therefore witnessing a double reification: that of the musical work, which is confused with a sound event as a physical reality, and that of the "hash", which is abstracted from its production process and considered as a homothetic portion of the work.

47However, this ideal of objectivity meets several limits. The first concerns the relevance of the part (the musical “fragment”) as a scale to judge a case of appropriation or duplication. Valérie Laure Benabou, a law expert for the French High Council of National Literary And Artistic Ownership, points out the risk that "the smallest parcel of a work [...] may surreptitiously become the representation of the whole", a process favoring "the extrapolation of the part towards the whole, losing sight of the qualifying unity of the work" (Benabou, 2014, p. 29). The definition of the work by sound may differ from other conceptions of the musical work, which would focus for example on the notions of lyrics, melody, or structure.

48The second limit is the setting of the matching ratio. If the matching ratio is relatively low, the risk is to detect a match between a video and a work from an amount of common data so limited that it leads to errors, some videos being mistakenly considered as cases of plagiarism35. If the correspondence rate is relatively high, the risk is the opposite: missing out on appropriations of segments of deposited musical works. The mathematizing of the originality judgment immediately calls for a political conception of copyright or neighboring rights that favors the circulation of cultural forms or, on the contrary, the control of public expression.

  • 36 Interview with the head of YouTube rights management for one of the main French digital music distr (...)
  • 37 Rights manager Rumblefish is said to have confirmed ownership of the song, before blaming YouTube f (...)

49The third limitation concerns the difference between the parameters and variables relevant to "isolate" a sound form and those relevant to judge the originality of a musical creation. Certainly, for several decades, it is the idea of music as sound which has dominated a diversity of musical aesthetics (Delalande, 2001). In some of them, a form of "sound identity" even seems to matter more than any other criterion to distinguish the originality of a creation: phenomena such as mixing or texture are part of the criteria of originality in popular music, and in particular in music genres such as techno or rap instrumentals. Hashing techniques also have to deal with difficulties related to the variety of uses of sound as a compositional and creative resource, such as the variability in the distribution of the most distinctive marks in the spectrogram: Kalenzenga takes the example of a soprano a capella, in which only the medium and high frequencies are relevant, while a rap or jazz piece will focus on low or medium bass frequencies. Ambient or noise music, which are precisely characterized by aesthetic criteria of “sonic identity”, rely on parts of the spectrum which seem poorly adapted to the coding by Content ID: as the tool is unable to clearly establish the singularity of the works in this repertoire, some record company representatives in charge of the daily management of online rights have given up using the tool for these genres36. In some cases, the problems of scale, threshold and frequency seem to add up: for example, some videos containing white noise or bird songs have been permanently blocked insofar as they were identified as corresponding to deposited works37.

50If Content ID seems to be designed for the detection of synchronization or sampling practices (the presence of similar sound fragments over a certain period of time in another musical work or in a video), it has been necessary to consider the extent to which it can take into account the diversity and evolution of tacit or explicit conventions at work in the diversity of musical cultures. Indeed, in current musical practices, the appropriation of particular aspects of musical form and sound, is either increasingly tolerated or, on the contrary, decried, depending on the cases, times and places.

51To illustrate tolerance, one can think of the use of “type beats” in rap music: these instrumentals are stereotyped sound signatures, very accurate imitations of other producers, or a producer’s variations on the same melodies, timbres and rhythms. For this very reason, their designers (the beatmakers) choose to sell them as a regular commodity, often renouncing to collect any royalties on their future uses, and bypassing any interest in identification or tracking tools38.

52On the other hand, taking a look at interpolation will show the evaluation of appropriation has been hardened. Interpolation, which has long been one of the stylistic figures of the classical repertoire, enjoyed some success in rap and electronic music in the 1980s, before once again becoming a major phenomenon in 2017, with a series of pop hits. This technique consists in re-recording part of a pre-existing song to integrate it into a new song. The new work can end up including several choruses or verses from other songs, which is then close to a medley. In his #1 hit “Shape of you”, Ed Sheeran uses the melody of another proven hit, “No Scrubs”, by female RnB group TLC. In “Closer”, The Chainsmoker uses a catchphrase from the 2015 hit “Over my head (Cable Car)” by The Fray in combination with five instrumental variations, each with a different timbre, and five vocal variations (male, female, pitch change).

53In standard industry practice, such interpolations usually need clearance with the original copyright holders before they can be used, although they are not “samples” per se. If an interpolation constitutes a "re-recording" which imitates its reference on several musicological parameters at the same time, are these parameters likely to cross a modeling based on the sound spectrum, and therefore to be the object of detections by Content ID to warn producers and rights holders of an "original"? Or will the importance given in the model to other parameters closer to the timbre be sufficient to exclude it?

54There is no public register of the “blocked” works to account for the way YouTube regulates the borders in this area, deciding what is legible appropriation and what violates the rules of originality. Interpolation is only one of the many variants that reveal the difficulty with rooting the objectivity of the musical work and its originality in a model of music and sound. Since most “blocking” happens pre-emptively and discreetly, it is hard to know precisely, as of now, which modeling techniques and arbitrations are specific to Content ID, and whether it enforces a flexible or hard policy in such cases.

55These situations show that as cultural productions are detected, and their managers dispossessed of part of their income for the benefit of hypothetical or real rights holders, what matters are the methods chosen to know and decide not only where the originality of a work lies, but what "music" consists in. It is an issue that can be formulated in terms of semiotic predilections (Jeanneret, 2008) in relation to music and works, or in relation to the grips thanks to which we take hold of them (Bessy and Chateauraynaud, 1995; Hennion, 2002;), or even with reference to the multiple "ontologies" within which they operate socially (Born, 2005). Depending on the genres, audiences and situations, music refers primarily to a performance, a written form, a materialized idea, a structure, a decipherable intention, the occasional variation of a collective style functioning as a "common good"... The objectification of the "sound form" does not always make it possible to do justice to these distinctions, which are at the heart of the different musical experiences, but also to the evaluation of what deserves to be protected by copyright and neighboring rights.

The (semi-)automation of judgment and the administration of the sensitive realm

56Judgments concerning originality are generally based not on a definition of what a musical work is in legal terms – since there seems to be a consensus about the objectification of authority and moral rights in music with regard to the media that fix the sound – but on what constitutes, in the work, an expressive form and how unique it is. Content ID's bias in favor of the sound fragment as an anchor of musical form and originality certainly corresponds to some of the aesthetic coordinates that have dominated popular music recorded since the 1950s, but this is not an obvious or indisputable reality.

57When it comes to defining what "originality" means, it could be argued that a melody, a rhythm or a timbre can take precedence over the "synthetic" unit composed by the variables of the spectrogram and the needs for efficiency in their algorithmic reduction; conversely, the automatic identification of a fragment can be opposed to the need to take into account the relationship of the detected data to an overall structure. The American legal tradition, despite precedents that may have discouraged sampling, has left this question open, as Siva Vaidhyanathan points out:

  • 39 Siva Vaidhyanathan (2003, pp. 117-118).

If copyright law is charged with protecting a particular arrangement of notes, should it protect the melody, the harmony, the rhythm, or all of the above? How long must that string of notes be to constitute a protectable segment of expression? Should music copyright law be most concerned with the “total concept and feel” of a protected work, or particular elements such as solos, riffs, or choruses?39

58Faced with the complex/stimulating issue of musical originality, American courts have developed a method based on two questions: whether or not the accused had access to the original work and a principle of "substantial similarity". These two criteria are supposed to be the basis to establish an intention to appropriate a work (and therefore the potential violation of a moral right), but jurisprudence has not always taken them into account. In 1976, the band The Chiffons brought a case to trial: the melody of George Harrison's song "My Sweet Lord" (1970) was nearly a measure by measure echo of their song "He's So Fine" (1963). At the trial, Judge Owen sentenced Harrison for being "unconsciously" influenced by the Chiffons' song, thereby bypassing the intentionality test.

  • 40 The ego-trip is a bragging performance, which is a rhetoric in its own right in certain styles of r (...)
  • 41 « When rapper Biz Markie sampled Gilbert O’Sullivan’s 1970s hit “Alone Again, Naturally”, the South (...)

59Sometimes, intentionality and access to the work are clear, but it is the question of "substantial similarity" that is treated differently. In 1991, rapper Biz Markie borrowed a short sample from Gilbert O'Sullivan's melancholy ballad "Alone Again" for a piece in the egotrip style40: Biz Markie was convicted and the judgment then confirmed the domination of moral rights over the principle of similarity. This trial had decisive consequences, since the use of samples in American rap has steadily declined since then. However, a few years later, the cover of the song "Pretty Woman" in a rap song by 2 Live Crew escaped conviction by being identified by the Supreme Court as a creative form of parody. Alex Sayf Cummings notes the difficulty to reconcile the two decisions: "Is one a "transformative appropriation" and the other a "theft appropriation"? (Cummings, 2013)41. One thing is certain: by YouTube's own admission, Content ID has no sense of humor and cannot make exceptions for parody covers, even though the American concept of fair use authorizes parody to promote freedom of expression and criticism.

60Intervening upstream from the judges' interpretation work and taking charge of the fatally aesthetic interpretation of appropriation, Content ID partially replaces it, even if it is ultimately the rights holders themselves who decide what constitutes legitimate appropriation. They can then apply their criteria, leaving the "appropriate" parties to challenge decisions until they manage to move beyond the contractual and technical logic of the system to bring judges back to court and obtain the release of blocked videos a process that can be long and costly.

61Other cases concern the presence of "incidental" music in videos that have been “captured live”: it may be involuntary, as seems to be the case in the very first YouTube video, "Me at the Zoo", by Jawed Karim (one of the three co-founders of the site), the legend of which indicates that it contains a fragment of Darude's "Sandstorm" trance classic, although it is not really striking in today’s available links to this YouTube video42. It may also be deliberate, as on the startling "Cecil Robert" channel, which features hits like Toto's "Africa", put through effects imitating the exhilarating acoustics of certain places in this case, a supermarket43.

  • 44 Some Internet users work to defeat the identification thresholds of Content ID by playing directly (...)

62With this type of video, as with the success of ASMR videos, which make the noise of everyday objects resonate, YouTube has become a space for the media coverage of an emerging sound culture that offers aesthetic experiences in unexpected sound or sensory forms. These experiences blur the typical definitions of music and indirectly renew the attempt of the historical avant-gardes to broaden the definition of musical practice towards experimental listening to objects or poetic listening to the sound environment (Saladin, 2007). In these conceptions, creative responsibility partly escapes the initiators of the sound productions fixed in supports: if everyone can become sensitive to environments and sound materials, it becomes difficult to know if their value lies in the artist's gesture, in the environment, in the material or in the listener's discriminant listening activity44.

63Unlike these experiments, the logic of Content ID consists in annexing, with the holders of controlled repertoires, the sound and the sensitive in a new legal and economic space. The sounds and music that circulate in the videos of Internet users, videographers and musicians are thus increasingly likely to be either "blocked" or integrated into "monetization" schemes.

Formats as "micropolitical" terrain

64Concretely speaking, the project to monetize a large number of music videos through advertising faces many difficulties. Videos that should escape the commercial domain are blocked (Calimaq, 2013) while legitimate uses of sound quotes, as in educational videos or parodies, are blocked as well, or their advertising revenues captured by publishing companies. Pierre-Carl Langlais argues that "false positives" should mechanically increase as more works are deposited in the system45. Finally, the rules authorizing the deposit of works favor the first to arrive in the system, i.e. the major record companies, those who are also in a position to have entire teams managing these new tools and the new revenues they are likely to generate46.

65YouTube recently confirmed the analyses outlined in this article by displaying on the Content ID help page the enigmatic condition that "reference content must be sufficiently distinct"47. Due to the structural and operational limitations of the Content ID tool, we are witnessing not only the extension of the scope of collection of remuneration attached to rights derived from copyright, but also the drawing of a new, more or less blurred line between music pieces ready to be transformed into digital goods and others that risk being excluded: the cases described above, including genres like ambient, drone and noise, as well as live versions or certain remixes which in practice are already excluded from the Content ID controls practiced by digital distributors48.

66When it comes to judging originality, Content ID constitutes an opacification since the calculation model is not public and a potential hardening of the jurisprudence, even if it is already relatively strict. While recognizing the theoretical need to maintain exceptions to the application of neighboring rights, the European Union nevertheless seemed to validate this same logic of governance by technology when it voted in favor of Article 13 of the Copyright Directive49. The increased responsibility of "new digital technology players" for the productions they circulate may lead a company like YouTube to make always more use of Content ID until it relies on automated filtering, which will have to work always more a priori in order to avoid any dispute.

67A mission note to the French government written by the President of the Higher Council Of Artistic Ownership (Japiot, 2019) is very clear on this subject:

Tools for the recognition of works on digital content sharing platforms are nowadays a crucial aspect both of respect for copyright and related rights and, through the agreements with right holders that they allow to set up, of remuneration for creative work. Article 13 of the proposed European Directive on Copyright in the Digital Marketplace should widen their range, by transforming these tools, set up on a voluntary basis, into mechanisms provided for and regulated by European Union law.

68In Germany, the vote on this directive has led to social movements stressing the risks of "pre-emptive censorship" that this legal-technical model presents (Reda, 2018). The French government is currently studying the possibility to apply these automated filtering methods to what it has labeled "hate content", suggesting the possible generalization of a technical governance model for online communication, operated by companies50.

69In these initiatives, what comes up as a priority from the emerging reflections and controversies concerns above all the balance of power between economic actors themselves (cultural industries versus "communication industries"), and then between these actors, political institutions and web users. When technology is questioned, it is about its efficiency: can it be precise enough in its filtering?

70The format-based approach developed in this article completes these discussions in several ways. It relativizes the promises that YouTube, one of the biggest players using this kind of filters, combines with automation, large-scale processing, and the computer processing that produces the "signature". It highlights the irreducible amount of decision-making, arbitration and axiology in the work of engineers who must establish variables and detection thresholds, arbitrate between the risks of "over- and under-detection", deal with the intrinsic entropy of modeling processes, and adopt psycho-acoustic and musicological standards that are not much questioned. It shows a discontinuity that the public discussion, held in terms of "application of copyright", fails to reveal: the move from a legal definition of the work based on an inquiry into the social significance of forms carried out by the legal institution, to a "sound ontology". Finally, it highlights the "archaeological" background that underlies these projects: the transformation of a legal tool for organizing the creative economy the copyright into an essentialist view of individual originality, "naturalized" in the individual ownership of works and its extension to the rights associated to the exploitation of phonograms (neighboring rights); the championing of technology and its supposed "objectivity", from the "fidelity" that feeds essentialist attitudes to the work in the form of the recording (aimed to disappear as the “medium”), to the confidence in the accuracy and effectiveness of identification and correspondence models.

71All these aspects reveal the existence of knowledge politics and potential conflicts of meanings nestled at the heart of the "micro" and infra-ordinary realities of media communication. The micro-political point of view on formats therefore raises four more global questions: how are cultural decisions embedded, routinized and made discreet in the designing of techniques? How can online aesthetics and creations resist projects to objectify culture and reveal their flaws? To what extent does communication from the professional worlds of web economics, law, music and IT contribute to revealing the differences between the two? To what extent does the communication of the professional worlds of web economics, law, music and computer science contribute to revealing the difficulties or limitations pertaining to rationalizing culture and controlling expression? Or, on the contrary, to reserving them for expert-only debates? Finally, to what extent can engineers open themselves to the plurality of relationships to culture and the realm of sensitivity, and musicians, mediators and guarantors of cultural expression appropriate the intelligence of the technical formats that support their activities?

72These questions need to be answered so that these formats, now invested by large-scale institutional, political and commercial projects, can be discussed and chosen, rejected or re-imagined by all.

Haut de page


Adorno, T. W. (1994). Introduction à la sociologie de la musique, Genève : Contrechamps.

Alías, F., Socoró, J., & Sevillano, X. (2016). A Review of Physical and Perceptual Feature Extraction Techniques for Speech, Music and Environmental Sounds. Applied Sciences, 6(5), 143.

Appadurai, A. (1986). The social life of things. Cambridge: Cambridge University Press.

Benabou, V. L. (2014). Rapport de la mission du CSPLA sur les œuvres transformatives. Paris : Conseil supérieur de la propriété littéraire et artistique.

Bessy, C., & Chateauraynaud, F. (1995). Experts et faussaires. Pour une sociologie de la perception. Paris : Métailié.

Beuscart, J.-S. (2006). La construction du marché de la musique en ligne : l’insertion économique et juridique des innovations de diffusion musicale en France. ENS Cachan.

Born, G. (2005). On Musical Mediation: Ontology, Technology and Creativity. TwentiethCentury Music, 2(1), 7-36.

Bouquillion, P., & Matthews, J.-T. (2010). Le Web collaboratif : mutations des industries de la culture et de la communication. Grenoble : Presses universitaires de Grenoble.

Bouquillion, P., Miège, B., & Morizet, C. (2006). À propos des mouvements récents (20042005) de concentration capitalistique dans les industries culturelles et médiatiques. Le Temps des médias, 6(1), 151-164.

Bruno, F., Lissovsky, M., & Ferraz Vidal Junior, I. (2018). Abstraction, expropriation, anticipation : note généalogique sur les visions machiniques de la gestualité. Réseaux, 211(5), 105.

Bullich, V. (2015). La propriété artistique comme régulation de la « trivialité ». Communication et langages, 185, 67-88.

Burkart, P., & McCourt, T. (2004). Infrastructure for the Celestial Jukebox. Popular Music, 23(3), 349-362.

Callon, M., Akrich, M., Dubuisson-Quellier, S., Grandclément, C., Hennion, A., Latour, B., Mallard, A., et al. (2013). Sociologie des agencements marchands. Textes choisis. Paris : Presses des Mines.

Cano, P., Batlle, E., Kalker, T., & Haitsma, J. (2005). A Review of Audio Fingerprinting. Journal of VLSI signal processing systems for signal, image and video technology, 41(3), 271-284.

Charrak, A. (2001). Raison et perception : fonder l’harmonie au XVIIIe siècle. Paris : Vrin.

Cummings, A. S. (2013). Democracy of sound: music piracy and the remaking of American copyright in the Twentieth century. Oxford: Oxford University Press.

Delalande, F. (2001). Le son des musiques, entre technologie et esthétique. Paris : Buchet-Chastel.

Eriksson, M. (à paraître). In Pursuit of Musical Identifications.

Everett, W. (2010). “If you’re gonna have a hit”: intratextual mixes and edits of pop recordings. Popular Music, 29(2), 229-250.

Fontanille, J. (1995). Sémiotique du visible. Des mondes de lumière. Paris : Presses universitaires de France.

Francastel, P. (1956). Art et technique. Paris : Gallimard.

Gomez Mejia, G. (2014). De quoi le « nuage » est-il le nom ? Le statut des supports face aux régimes du cloud computing. Communication & langages, 182(4), 77-93.

Goyet, S. (2017). De briques et de blocs. La fonction éditoriale des interfaces de programmation (API) web : entre science combinatoire et industrie du texte. Thèse de doctorat, Celsa Sorbonne Université.

Gras, S.-E. (2014). L’écoute en ligne : Figures du sujet écoutant et mutations des espaces musicaux sur Internet. Thèse de doctorat, Celsa Sorbonne Université.

Heald, P. J. (2014). How Notice-and-Takedown Regimes Create Markets for Music on YouTube: An Empirical Study. SSRN Electronic Journal. From abstract=2416519

Hennion, A. (2002). Une pragmatique de la musique : expériences d’écoutes. Petit retour arrière sur le séminaire Aimer la musique. MEI. Media et information, 17, 31-43.

Heuguet, G. (2018). Métamorphoses de la musique et capitalisme médiatique. Au prisme de YouTube (2005-2018). Thèse de doctorat, Celsa Sorbonne Université.

Japiot, O. (2019). Lettre de mission CSPLA Hadopi et CNC. Paris : Conseil supérieur de la propriété artistique.

Jeanneret, Y. (2001). Autre chose qu’un discours, davantage qu’un accompagnement, mieux qu’une résistance. Terminal, 107-117.

Jeanneret, Y. (2007). Y a-t-il (vraiment) des technologies de l’information ? Villeneuve-d’Ascq : Presses universitaires du Septentrion.

Jeanneret, Y. (2008). Penser la trivialité. Paris : Lavoisier.

Jeanneret, Y. (2011). Complexité de la notion de trace : de la traque au tracé. In B. Galinon-Mélenec (dir.), L’homme trace : perspectives anthropologiques des traces contemporaines. Paris : CNRS Éditions.

Jeanneret, Y. (2014). Critique de la trivialité : les médiations de la communication, enjeu de pouvoir. Paris : Éditions Non standard.

Jeanneret, Y., & Souchier, E. (2002). La communication médiatisée est-elle un « usage » ? Communication & langages, 132(1), 5-27.

Kaltenecker, M. (2010). L’oreille divisée : les discours sur l’écoute musicale aux XVIIIe et XIXe siècles. Paris : Éditions Musica Falsa.

Kim, J. (2012). The institutionalization of YouTube From user-generated content to professionally generated content. Media, Culture & Society, 34(1), 53-67.

Kittler, F. (2018). Gramophone, film, typewriter. Dijon : Les Presses du Réel.

Latour, B. (1996). Petites leçons de sociologie des sciences. Paris : Seuil.

Latour, B. (2005). La science en action : introduction à la sociologie des sciences. Paris : La Découverte.

Mackenzie, A. (2006). Codecs: Encoding/decoding images and sounds. From

McIntosh, H. (2016). Vevo and the Business of Online Music Video Distribution. Popular Music and Society, 39(5), 487-500.

Morizot, J. (2015). Esthétique et naturalisation. Nouvelle revue d’esthétique, 15(1), 11.

Mulligan, M. (2015). Awakening: The music industry in the digital age. London: MIDiA Research.

Nikoltchev, S. (2007). Legal aspects of video on demand. Strasbourg: European Audiovisual Observatory.

Reda, J. (2018). Censorship Machines. Article 13 in conjunction with Recitals 38 and 39 of the proposed EU copyright reform/expansion. Bruxelles : Commission européenne.

Saladin, M. (2007). Les musiques « expérimentales », une polémique en devenir. Audimat, 7, 186-207.

Smythe, D. W. (1981). On the audience commodity and its work. Media and Cultural Studies: Keyworks, 230-256.

Srnicek, N. (2018). Capitalisme de plateforme. Montréal : Lux.

Sterne, J. (2018). MP3. Économie politique de la compression. Paris : La Rue Musicale.

Sterne, J. (2015). Une histoire de la modernité sonore. Paris : La Rue Musicale/La Découverte.

Théberge, P. (1997). Any sound you can imagine: making music/consuming technology. Middletown, Connecticut: Wesleyan University Press.

Vaidhyanathan, S. (2003). Copyrights and copywrongs: the rise of intellectual property and how it threatens creativity. New York: New York University Press.

Valle, A. (2015). Towards a Semiotics of the Audible. Signata. Annales des sémiotiques/ Annals of Semiotics, 6, 65-89.

Haut de page


1 Carpentier, L., (Retrieved October 31, 2013).

2 For a genealogy of this construction in France, cf. Beuscart (2006).

3 The following analysis uses english-langage documentation regarding US-based policies, while sources on copyright are coming from both US and European contributions.

4 The "theatrophone" was supposed to allow, over a century before the streaming boom, listening to operas at home via the phone, in exchange for a package.

5 Haskins, C., The Outline. (Retrieved April 9th, 2019).

6 Introduced by law n ° 85-660 of July 3, 1985 in France, neighboring rights extend rights of authorization and remuneration on the use and exploitation of works to entities other than songwriters, such as the performers but also the phonogram editors (or producers). In some cases (self-production), these entities may correspond to the same person, but it is often the record companies, as producers, who collect and collect neighboring rights. These revenues can be shared with the artist in case of co-production, or not. In the rest of this article, the reference to "copyright" should not be understood in the strict sense of the rights of songwriters, but refers to the overall legal framework which allows the collection of rights associated with the production of phonograms.

7 Law n° 2004-575 of June 21, 2004 « Pour la confiance dans l’économie numérique ».

8 Harvey, E., (Retrieved April 9th, 2019).

9 « To clear up confusion, this is not a new feature. The dispute process has been in place since Content ID first launched in October 2007. We’ve changed some text to make that clear » (« Content ID & Fair Use » 2010).

10 The YouTube Team, YouTube Blog. (Retrieved April 12th, 2017).

11 YouTube, On line: (Retrieved December 15 2013).

12 The European directive on digital commerce (applied in France), designed to promote the competitiveness of European cultural industries, is less tolerant than the DMCA in terms of exceptions to copyright, since it does not include exceptions from parody or information.

13 Walk, H., (Retrieved September 25th 2014).

14 Walk, H., (Retrieved September 25th 2014).

15 Founded by law professor Lawrence Lessig, Creative Commons is an American "non-profit" organization that has published several copyright licenses. Creative Commons licenses allow authors to specify the rights they wish to exercise and those they renounce in relation to a given work, for example by authorizing non-commercial reuse in advance without negotiation on a case-by-case basis (

16 Calimaq, S.I.Lex. On line: (consulté le December 23rd, 2013).

17 Audible Magic, On line: (July 9th, 2018).

18 The circulation of decoy MP3 files, where an artist name and a title in the file name covered an inaudible or totally different piece, constituted a privileged strategy of the majors in their fights against the circulation of music on the peer-to-peer software.

19 Blum, T. L. S. F., Keislar, D. F. B., Wheaton, J. A. F. et Wold, E. H. E. C., Google Patents. On line: (Retrieved April 9th, 2019).

20 The source code indicates "a set of texts or resources allowing the execution of a program or a web page. Each text contains a certain number of instructions given to the machine, written according to the syntax of a programming language" (Goyet, 2017). Displaying part of the source code for pages on a website is one of the common options in most web browsing software available on the market.

21 TorrentFreak, On line: (Retrieved June 25th, 2018).

22 Pace, M. R. et Talley, B. C., Google Patents. On line: (consulté le 9 avril 2019).

23 Cf. Ecklund, R., Audio Watermarking Techniques. On line: (consulté le 9 avril 2019).

24 YouTube, YouTube Help. (Retrieved April 3rd, 2019).

25 YouTube, YouTube Help. On line: (Retrieved April 3rd 2019).

26 Langlais, P.-C., Sciences communes. (Retrieved September 20th, 2016).

27 For an exception concerning the genealogical approach to machine vision, and whose results on the "dehumanization" of the definition of the modeled object converge with our own reflections, see Bruno, Lissovsky and Ferraz Vidal Junior (2018).

28 The cepstrum is the transformation of a mathematical modeling of temporality into another mathematical domain for its treatment; it uses in particular the amplitude characteristics of the signal.

29 Wang, A. L.-C. et Smith III, J. O., Google Patents. (Retrieved April 19  2019).

30 Kalenzaga, C., Coding Geek. (Retrieved August 6th, 2015).

31 Segers, J., (Retrieved April 3rd 2019).

32 On the criticism of the "averaging" of hearing by laboratory tests, cf. Tern (2018).

33 Jovanovic, J., Toptal. (Retrieved March 30 2019).

34 Interview with the head of YouTube rights management for one of the main French digital music distributors, March 2019.

35 Langlais, P.-C., Sciences communes. (September 20th, 2016).

36 Interview with the head of YouTube rights management for one of the main French digital music distributors, March 2019.

37 Rights manager Rumblefish is said to have confirmed ownership of the song, before blaming YouTube for the blocking. O’Neill, M., April 3rd, 2018).

38 Setaro, S. (2019). (Retrieved March 30th, 2019).

39 Siva Vaidhyanathan (2003, pp. 117-118).

40 The ego-trip is a bragging performance, which is a rhetoric in its own right in certain styles of rap music.

41 « When rapper Biz Markie sampled Gilbert O’Sullivan’s 1970s hit “Alone Again, Naturally”, the Southern District Court of New York held him liable for copyright infringement, while 2 Live Crew found that their use of Roy Orbison’s “Pretty Woman” in a raunchy rap song qualified as a creative form of parody in a landmark 1994 Supreme Court case. Is one a “transformative appropriation”, and the other one theft? The difference between the two cases shows the sometimes arbitrary nature of copyright. »

42 Jawed, (Retrieved August 18th, 2018).

43 Cecil Robert, (August 18th, 2018).

44 Some Internet users work to defeat the identification thresholds of Content ID by playing directly on the computer processing of the song, for example by playing on the speed of reading the recording: to speed up 'a slowed-down piece using the options in the player software allows to discover a piece thus passed under the radar of Content ID.

45 Langlais, P.-C., Sciences communes. (September 20th, 2016).

46 YouTube, YouTube Help. (Retrieved April 3rd, 2019).

47 YouTube, YouTube Help. (Retrieved April 8th, 2019).

48 Interview with the head of YouTube rights management for one of the main French digital music distributors, March 2019.

49 Signoret, P., Le (Retrieved July 9th, 2018).

50 Champeau, G., (Retrieved January 21st, 2016).

Haut de page

Table des illustrations

Titre Figure 1. Examples of two "hashes" (for an original image and its thumbnail) represented as a digital key as proposed by the engineer Jens Segers31
Crédits Segers, J.,
Fichier image/jpeg, 11k
Crédits Segers, 2014
Fichier image/jpeg, 122k
Titre Figure 3a. Two figurations of the mediations involved in the process of producing a "perceptual hash" - figuration by Cano et al.
Crédits Source: Cano, Batlle, Kalker & Haitsma, 2005
Fichier image/jpeg, 49k
Titre Figure 3b. Figuration de Jovanovic33
Crédits Source: Jovanovic, 2014
Fichier image/jpeg, 22k
Haut de page

Pour citer cet article

Référence électronique

Guillaume Heuguet, « Towards a micropolitics of formats »Revue d’anthropologie des connaissances [En ligne], 13-3 | 2019, mis en ligne le 01 septembre 2019, consulté le 02 avril 2020. URL :

Haut de page


Guillaume Heuguet

Researcher at GRIPIC (Sorbonne University), where he runs the Transnum program (Thinking digital as transformation), and associate researcher at IRMECCEN (Sorbonne Nouvelle), he is the author of an anthology of popular music studies published by La Rue Musicale and the editor-in-chief of the music critic journal Audimat.

Address: GRIPIC, Celsa Sorbonne Université, 77 rue de Villiers, FR-92200 Neuilly-sur-Seine (France).
Email: guillaume[at]

Haut de page

Droits d’auteur

Licence Creative Commons
Les contenus de la Revue d’anthropologie des connaissances sont mis à disposition selon les termes de la Licence Creative Commons Attribution - Pas d'Utilisation Commerciale - Pas de Modification 4.0 International.

Haut de page
  • Logo Société d’anthropologie des connaissances
  • Logo DOAJ – Directory of Open Access Journals
  • Logo ERIH PLUS | NSD
  • OpenEdition Journals