AI in the Music Industry – Part 17: ‘Fake Drake’ and the Problem of Deep Fakes

In mid-April 2023, there was a lot of excitement when the song “Heart on My Sleeve” appeared on the Internet and went viral. It features Canadian rap superstar Drake in a duet with his famous compatriot The Weeknd. Fans were already thinking that Drake had released a new song with a crazy new promotional campaign when it emerged that his voice had been artificially imitated by someone disguised in a sheet and sunglasses, calling himself ghostwriter977, who then posted the song on TikTok.[1] The song was already available on Spotify on 4 April[2] and then also appeared on other streaming portals and YouTube.[3]

‘Fake Drake’, as the AI-generated song was soon dubbed, not only triggered a worldwide media frenzy, but also brought the rights holders, and particularly the labels, to the barricades. Universal Music Group, which has both Drake and The Weeknd signed to its Republic sub-label, was quick to voice its disapproval: “[T]he training of generative AI using our artists’ music (which represents both a breach of our agreements and a violation of copyright law) as well as the availability of infringing content created with generative AI on DSPs, begs the question as to which side of history all stakeholders in the music ecosystem want to be on: the side of artists, fans and human creative expression, or on the side of deep fakes, fraud and denying artists their due compensation.”[4]

Universal had already sent an email to the music streaming services before the ‘Fake Dake’ song emerged in March 2023, in which the company threatened AI-generated music fakes: “We will not hesitate to take steps to protect our rights and those of our artists.”[5] This was effective. After ‘Heart on My Sleeve’ had been shared 9 million times on TikTok and generated 250,000 streams on Spotify within two weeks, and the YouTube video had been clicked almost 200,000 times in two days, the streaming portals removed the track on the evening of 17 April, and YouTube suspended ghostwriter977’s account.[6]

However, the ‘Fake Drake’ song is just the tip of an iceberg of deep fakes that can be found on the Internet. Click through YouTube and you will find numerous voice clones of current and former US presidents engaged in rap battles.[7] Or you can smile at “Always Look on the Bright Side of Life” from the classic Monty Python film “The Life of Brian”, interpreted by the choir of the North Korean People’s Liberation Army.[8] This was produced by Dustin Ballard,[9] who allows himself numerous entertaining musical AI jokes on his YouTube channel with the programmatic title “There I Ruined It”. An AI-generated voice of Bruce Springsteen interprets his global hit “Born in the U.S.A.” in bossa nova style[10] or you can find a heavy metal version of Taylor Swift’s “Shake It Off”.[11] However, by the beginning of 2024, the videos are no longer available on the YouTube channel. Instead, the message reads: “Unfortunately, YouTube makes it nearly impossible for creators to assert the ‘Fair Use’ defense against record labels and publishers who automatically reject any appeals.”[12]

Nevertheless, since the beginning of 2023, there has been a boom in the creation of generative voice AI applications. Since January 2023, ElevenLabs, founded in 2022 by former Google and Palantir employees,[13] has been offering an AI voice imitation app that can be trained not only on your own voice, but also on the voices of other people and, of course, singers. Voice clones can be created in 29 languages and more than 50 accents.[14] The Cambridge-based company Suno AI[15] is pursuing a similar objective with its text-to-audio generator ‘Bark’, which can be used to create your own voice, as well as music and other sound files in an AI format. Audio files can currently be output in 13 languages with their respective accents.[16]

Another popular application is Voicify.ai. The company was founded in 2022 by 20-year-old Aditya Bansal, then a computer science student at the University of Southampton, right from his dorm room. In an interview with the Financial Times, the young man, dressed in Nike sneakers, shorts and a T-shirt, sits relaxed in front of his dormitory bed with his laptop on his knees, and in another photo is seen looking over his shoulder as he works on the computer.[17] The Financial Times told Bansal that he could make good money from cover versions of the AI voice clones. He has also been approached by record labels to create voice models of their artists for demo recordings, as a kind of sketch for the subsequent recording process.

However, the student will also receive mail from the Recording Association of America (RIAA), which added Voicify.ai to its list of copyright infringers, according to a report dated 6 October 2023. The RIAA has introduced a new category of music copyright infringement called ‘AI Vocal Cloning’ and Voicify.ai is the only service mentioned by name. The report shows that 8.8 million users used Voicify to create voice clones.[18] The US music industry association complains that 2023 has brought an eruption of AI voice cloning services, “(…) that infringe not only the rights of the artists whose voices are being cloned but also the rights of those that own the sound recordings in each underlying musical track. This has led to an explosion of unauthorized derivative works of our members’ sound recordings which harm sound recording artists and copyright owners.”[19] The RIAA also explains how copyright infringement occurs from their perspective: “The service stream-rips the YouTube video selected by the user, copies the acapella from the track, modifies the acapella using the AI vocal model, and then provides to the user unauthorized copies of the modified a capella stem, the underlying instrumental bed, and the modified remixed recording.”[20]

In February 2024, the British Phonograph Industry (BPI) also sent a letter through its lawyers to Voicify, which has renamed itself Jammable, threatening legal action if the company did not stop infringing copyright through voice cloning.[21]

Therefore, voice clones of famous persons such as Donald Trump, Barack Obama, Cristiano Ronaldo or Angela Merkel, but also singers such as BTS-Stars Jeon Jungkook and Suga, Elvis Presley, Bruno Mars, Taylor Swift, Justin Bieber can be made on the Jammable[22] webpage to generate new songs.

According to the RIAA, Jammable/Viocify violates both the moral rights of musicians and the master rights (neighbouring rights) in recordings. In his article for Music Business Worldwide, Daniel Tencer points out that international copyright law has harmonised exploitation rights, but not the moral rights of artists. In the US, for example, only 19 out of 50 states – including California, New York and Florida – have statutory publicity rights.[23] Above all, it is about the right of publicity, which protects against the unlawful appropriation of a person’s name, image or other identity features, such as their voice, for commercial purposes.[24] There is also judicial decision that predates the digitalisation of music. In 1988, the US actress and singer Bette Midler sued the car manufacturer Ford for using a voice impersonator in a series of advertisements who imitated Bette Midler without her consent. The Court of Appeal ruled that Midler’s voice was part of her personality and therefore distinctive. It is therefore unlawful to imitate her voice without her express consent and authorisation.[25]

Based on this idea, Universal Music Group’s submission to the US Copyright Office calls for the creation of a national right of publicity in the US, similar to that of some states, to protect a person’s name, image and voice in order to prevent the unauthorised use of voices in voice cloning.[26]

Plaintiffs in the US would therefore have a starting point to take action against AI-generated voice clones, especially if there is commercial use. This is certainly the case with Jammable/Viocify, as the website offers three monthly subscription models. For €1.99 per month, users can create 25 credits- i.e. sound files. For €9.99 per month, there is no limit to the number of credits you can create and you can also create three custom models, which means that each voice can be converted into a different model. For €89.99 per month, the number of custom models you can create increases to ten.[27]

The use of copyrighted music for voice cloning could also be a problem for Jammable/Voicify. Since an audio file has to be uploaded or a YouTube link has to be provided in order to use the application, it is in any case a reproduction of a copyrighted music recording. This is a criminal offence in all jurisdictions with functioning copyright laws. Even if the copyright infringement is committed by the user, Jammable/Viocify could at least be considered a contributor.

The legal situation with the ‘Fake Drake’ song is more complicated. After all, ‘Heart on My Sleeve’ is an original composition performed by AI-generated voices that like Drake and The Weeknd. So it is clear that the copyright and related rights in this case belong to the person with a sheet over his head and sunglasses on, whoever that may be. But it is still a deep fake. In their article “Regulating Deep-Fakes: Legal and Ethical Considerations,” Edvinas Meskys et al. provide four case studies in which deep fakes could have legal consequences.[28] In this particular case, it is a creative deep fake. AI is used to create new content that is a new form of creative expression. Since ‘Heart on My Sleeve’ was obviously created and distributed in the US, its legal standards apply. If the rights holders go to court, they would have to prove that the deep fake is an infringement of their copyright. This should be possible in this particular case. Although the AI was trained on millions of sound files, in the case of ‘Heart on My Sleeve’ it is clear that the music recordings of Drake and The Weeknd, which are clearly protected by copyright, were used. For example, the song contains a sample by music producer Metro Boomin (“If young Metro don’t trust you, I’m gon’ shoot you”), which should have cleared with the rights holder.[29] However, the interesting question is whether the copyright infringement was committed by ghostwriter977 by creating a specific recording or whether the copyright was already infringed in advance by training the AI. In the latter case, any training of an AI with copyright-protected data would be an infringement, which would have far-reaching consequences for the use of AI and could lead to it being banned.

Let’s assume that the creator of the specific song is sued for copyright infringement. The defendant could now refer to the ‘fair use’ doctrine [30] beziehen, which, according to Section 107 of the US Copyright Act, states that copyrighted works may be used for the purposes of criticism, comment, reporting, teaching, scholarship and research without the consent of the copyright holder. The court would then have to carry out a case-by-case examination in which the four-stage test is applied to determine whether ‘fair use’ within the meaning of the law exists.

The first step is to examine the purpose and nature of the use of the work, distinguishing between commercial use and use for educational purposes without the intention of making a profit. The defendant would have to prove that the song was created with non-commercial intent and fair moral motives, e.g. for the purpose of parody.

In a second step, the type of copyright-protected work must be included in the analysis. Specifically, we are talking about copyright-protected music recordings by Drake and The Weeknd and the number of works used in AI training could be narrowed down. However, the greater the number of input data, the more difficult it is to establish a causal link to the incriminated output.

The third aspect of the “fair use” doctrine is the extent of the portion of the protected work used and its substantive relevance. Since there are virtually no recognisable parts of the source material in ‘Heart on My Sleeve’, the interference appears to be insignificant. The quality of the interference results from the recognisability of the voices, and this could certainly be assessed as substantial.

Finally, the economic impact of the deep fake on the market value of the copyrighted material used must also be taken into account. A negative influence of “Heart on My Sleeve” on the songs of Drake and The Weeknd will not be so easy to prove. There could even be a positive PR effect for their songs.

The interpretation of the “fair use” provision regarding the ‘Fake Drake’ song is of course my personal judgement and does not represent a valid legal opinion. Therefore, the rights holders would have to go to court and sue for copyright infringement. A court would then have to decide whether there was any unlawful interference at all and whether ‘fair use’ could be claimed under the law. In my very subjective opinion, it will be very difficult for defendants to successfully argue ‘fair use’.

In the US and in common law countries, successful action could be taken against deep fakes such as ‘Heart on My Sleeve’ on the basis of ‘fair use’ provisions. However, in countries that follow the continental European legal tradition, such as the European Union, copyright law offers little to no protection against deep fakes that imitate the voice of living or deceased musicians in music recordings. Although Art. 7 of the Rome Treaty of 1961 already granted performers[31] the exclusive right to prohibit the broadcasting and communication to the public of their performance and the reproduction of the fixation of their performance under certain conditions, this right of personality does not protect against deep fakes. The WIPO Performances and Phonograms Treaty (WPPT) of 1996[32] strengthened the economic rights of performers (Art. 7 to 10) and extended their personal rights to include the obligation to give their name (Art. 5.1), but even this does not provide a means of combating AI voice imitation. It was not until the Beijing Treaty (2012) that not only the economic powers were reaffirmed, but above all the personal rights of performers were strengthened. In addition to attribution, as already provided for in the WPPT, performers are now granted the right to object to any distortion, mutilation or other alteration of their audiovisually fixed performance that would damage their reputation.[33] The problem is that the EU, like the US and the UK, has signed the Beijing treaty but not yet ratified it.[34]

Nevertheless, many countries in the European Union have codified moral rights that could also protect against unwanted deep fakes à la ‘Heart on My Sleeve’. In Germany e.g., this is primarily protection against the distortion of a work under Section 14 of the Copyright Act. This paragraph states that every author has the right “to prohibit any distortion or other impairment of his work that is likely to jeopardise his legitimate intellectual or personal interests in the work [translation by the author]”.[35] The aforementioned legal standard therefore shows that both authors and performers are protected against serious counterfeiting of their intellectual performances. Similar protection mechanisms can also be found in the copyright laws of other EU Member States. It remains to be seen whether these copyright provisions are sufficient for successful legal action against deep fakes. It would therefore be desirable to clarify the protection of the intellectual rights of authors and performers against AI-generated deep fakes on the basis of the Beijing Treaty.

However, there is another way in which rights holders can take legal action against deep fakes. Universal Music Group hinted at this in its public financial report (earnings call) for the first quarter of 2023. The Vice President of the group, Michael Nash, explains with reference to the “Fake Drake” song: “Soundalikes which serve to confuse the public as to the source or origin, or which constitute a commercial appropriation of likeness in the form of a distinctive voice, are all clearly illegal.”[36] The universal representative thus refers to legal regulations against unfair competition that can be found in virtually every legal system. Specifically, in Germany this is the Unfair Competition Act (UWG),[37] in which §4 no. 3 protects against unauthorised imitation. Accordingly, anyone who imitates a competitor’s goods and services and thereby deceives customers is acting unfairly and is in breach of the law. This legal standard could be applied to counterfeiters if they are competitors in a market. However, it would not apply to individuals who create deep fakes with the intent to deceive, but only to the companies that create and market the counterfeit products. The voice cloning platforms could therefore defend themselves by claiming that they provide the tools to create the deep fakes but are not themselves involved in the dishonest act. In the absence of relevant case law, it is currently unclear whether unfair competition laws provide effective protection against deep fakes.

---

Endnotes

[1] See TikTok/@ghostwriter977, https://www.tiktok.com/@ghostwriter977/video/7222027667132960046, “Video currently not available”, accessed: 2024-05-27.

[2] Rolling Stone, “Viral Drake and The Weeknd AI Collaboration Pulled From Apple, Spotify”, April 17, 2023, accessed: 2024-05-27.

[3] YouTube, “ghostwriter – heart on my sleeve (Drake x The Weeknd AI) Official Audio”, May 31, 2023, “Video currently not available”, accessed: 2024-05-27.

[4] Cited in: Music Business Worldwide, “Universal Music Group responds to ‘fake Drake’ AI track: Streaming platforms have ‘a fundamental responsibility to prevent the use of their services in ways that harm artists'”, April 17, 2023, accessed: 2024-05-27.

[5] Cited in: Financial Times, “Streaming services urged to clamp down on AI-generated music”, April 12, 2023, accessed: 2024-05-27.

[6] Music Business Worldwide, “This AI Drake rip-off already has 250,000 plays on Spotify. How will the music industry respond?”, April 17, 2023, accessed: 2024-05-27.

[7] YouTube user ‘juanjo_sound’ has created a number of celebrity rap battles using AI, with both the voices and lyrics generated by AI, including Donald Trump vs. Joe Biden, accessed: 2024-05-27.

[8] YouTube, “Always Look on the Bright Side of Life – North Korean Edition”, March 19, 2022, accessed: 2024-05-27.

[9] On his homepage (https://www.dustincreative.com/), Ballard introduces himself as Creative Group Head of the Dallas-based advertising agency ‘The Richards Group’, who is happily married and has two small children. In his spare time, he runs the YouTube channel “There I Ruined It”, subscribed to by more than 700,000 YouTube users.

[10] YouTube, “Born in the U.S.A. – Bossa Nova Edition”, April 13, 2021, accessed: 2023-10-16 (currently not available).

[11] YouTube, “Shake It Off (Taylor Swift) – Heavy Metal Edition”, 24. August 2020, accessed: 2023-10-16 (currently not available).

[12] YouTube, “There I Ruined It”, accessed: 2024-05-27.

[13] ElevenLabs, “About ElevenLabs”, n.d., accessed: 2024-05-27.

[14] ElevenLabs, “High Quality AI Voice Cloning”, n.d., accessed: 2024-05-27.

[15] Suno AI, “About Suno”, n.d., accessed: 2024-05-27.

[16] Heise.de, “Audio-KI ‘Bark’ erzeugt natürliche Sprache und kann sogar singen”, April 24, 2023, accessed: 2024-05-27.

[17] Financial Times, “Can AI make me a musical star?”, May 27, 2023, accessed: 2024-05-27.

[18] Recording Association of America (RIAA), 2023, RIAA Submission to Comment Request for the 2023 Review of Notorious Markets for Counterfeiting and Piracy, RIAA: Washington D.C., p 14.

[19] Ibid.

[20] Ibid.

[21] Music Business Worldwide, “AI vocal cloning app Voicify offers 3,000 deepfake models to replicate artists’ voices. Now it faces legal action from the UK’s music industry”, March 18, 2024, accessed: 2024-05-27.

[22] Jammable, “Models”, accessed: 2024-05-27.

[23] In January 2024, a bipartisan bill in the US House of Representatives called the “No Artificial Intelligence Fake Replicas And Unauthorised Duplications Act”, aptly abbreviated as the “No AI FRAUD Act”, seeks to establish a nationwide “right of publicity”. Se Music Business Worldwide, “Music industry applauds introduction of ‘No AI FRAUD Act’ in US Congress”, January 10, 2024, accessed: 2024-05-27 and Music Business Worldwide, “Major record companies hate AI voice-cloning platforms that don’t pay. The one they hate most was created by a 20-year-old UK student”, October 12, 2023, accessed: 2024-05-27.

[24] International Trademark Association, “Right to Publicity”, n.d., accessed: 2024-05-27.

[25] United States Court of Appeals for the Ninth Circuit, Bette Midler v. Ford Motor Company, 849 F. 2d 460, June 22, 1988.

[26] Universal Music Group, 2023, Notice of Inquiry “Artificial Intelligence and Copyright” before the United States Copyright Office, Docket No. 2023-6, p 94.

[27] Jammable, “Pricing”, accessed: 2024-05-27.

[28] Edvinas Meskeys, Aidas Liaudanskas, Julija Kalpokiene und Paulius Jurcys, 2020, “Regulating Deep-Fakes: Legal and Ethical Considerations”, Journal of Intellectual Property Law & Practice, vol 15(1), pp 4–11.

[29] Music Business Worldwide, “Universal Music Group: Yes, ripping off Drake’s voice for that AI track was illegal – and we’re certain of it”, April 27, 2023, accessed: 2024-05-27.

[30] US Copyright Act, 17 U.S.C. 107 – Limitations on exclusive rights: Fair use.

[31] This includes “actors, singers, musicians, dancers and other persons who perform, sing, recite, read, play or in any other way present literary or artistic works”, as specified in Article 3 of the Rome Treaty.

[32] The WIPO Performances and Phonograms Treaty was incorporated into the European Commission’s body of law in 2000. Official Journal of the European Communities No. L 089 of 11/04/2000 S. 0015 – 0023, http://data.europa.eu/eli/agree_internation/2000/278(2)/oj, accessed: 2024-05-27.

[33] Beijing Treaty on Audiovisual Performances of the WIPO of 24 June 2012, TRT/BEIJING/001, accessed: 2024-05-27.

[34] An updated list of WIPO contracting parties that have signed and ratified the Beijing Treaty can be found on the WIPO website: WIPO-Administered Treaties – Beijing Treaty on Audiovisual Performances, accessed: 2024-05-27.

[35] § 14 Entstellung des Werkes, Gesetz über Urheberrecht und verwandte Schutzrechte (Urheberrechtsgesetz) of the Federal Republic of Germany, accessed: 2024-05-27.

[36] Music Business Worldwide, “Universal Music Group: Yes, ripping off Drake’s voice for that AI track was illegal – and we’re certain of it”, April 27, 2023, accessed: 2024-05-27.

[37] Federal Republic of Germany, Gesetz gegen unlauteren Wettbewerb (UWG).