Source: Music Pioneer
Image source: Generated by Unbounded AI
Recently, Google has launched a series of initiatives around AIGC music.
On November 16, DeepMind, a subsidiary of Google, released its latest AIGC music generation model Lyria and collaborated with YouTube to create two important application scenarios: Dream Track and Music AI tools. At the same time, Google and YouTube have extended the AI recognition tool SynthID to the music field to watermark the AI works generated by Lyria.
In addition to the various AI music tools that have been introduced, Google is also actively addressing the copyright and ethical issues facing AIGC music. On November 14, YouTube, a subsidiary of Google, also released a blog post titled "Approaches to Responsible AI Innovation on YouTube," announcing a series of governance rules for AIGC content that will be gradually implemented in the coming months.
It can be seen that Google is rapidly laying out its position in the AIGC music field with its powerful music production tools and clear governance path. On the other hand, major copyright holders such as Universal Music and Warner Music, as well as platform providers like Spotify and Deezer, are also not idle, and the midfield battle of AIGC music has already begun.
Lyria, Currently the Most Powerful AIGC Music Tool
"Changing the future of music creation."
On November 16, DeepMind released a blog post with this concise title, introducing their latest AIGC music generation model Lyria. Compared to AI-generated dialogue audio, AI music models need to "simultaneously generate long sequences of sound," and the difficulty lies in "maintaining the continuity of the music in terms of perception."
This new AI model is more mature and excels in "generating high-quality music with instruments and vocals, completing tasks such as transforming different music styles and writing song continuations, allowing users to fine-tune the style and performance of the output music in more detail."
At the same time, DeepMind collaborated with YouTube to create two important application scenarios for Lyria: Dream Track and Music AI tools.
Among them, Dream Track is mainly aimed at short video creators on YouTube Shorts. This tool currently provides a series of authorized artist sound sources to some American creators. By simply entering a theme and selecting an artist, the Lyria model will generate a 30-second video soundtrack and simultaneously generate the soundtrack and lyrics.
As of now, the participating artist sound sources include Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Sia, T-Pain, Troye Sivan, Papoose, and more.
Music AI tools are mainly aimed at musicians, artists, and producers. Google stated that their researchers have been collaborating with artists, songwriters, and producers in YouTube's AI incubator to test this tool in order to better support music creation.
On Music AI tools, musicians can generate corresponding scores by humming a melody, generate AI singing by inputting a MIDI chord, and freely add a section of instrumental accompaniment in the vocal track to generate corresponding music works.
With this tool, musicians can create original music works with vocals and instruments, and then AI will help them experiment with various music styles and fine-tune complex accompaniments, ultimately creating a complete work that includes instrumental accompaniment and vocals.
In addition, to protect the rights of musicians and creators, all music works created and released with the assistance of the Lyria model will be marked and identified using SynthID. In November, SynthID was officially expanded to add and recognize AI watermarks in the music and audio field.
It is understood that SynthID will embed watermarks in AIGC audio content, which are inaudible to the human ear and do not affect the user's listening experience. Only when the sound is converted into a two-dimensional visualized spectrogram, will the watermark be captured by recognition tools. Even if the song undergoes noise addition, MP3 compression, acceleration, or deceleration, SynthID can still detect the presence of the watermark in the song.
Upon its release, Lyria has received widespread attention for its powerful features, and foreign media have even referred to it as "the future of the music industry." However, the powerful Lyria is built on the technical foundation accumulated by Google in its previous practices in AudioLM and MusicLM.
AudioLM is an AI music model released by Google in September 2022. As a pure audio model, it can generate and suggest coherent music styles simply by listening to audio.
In February 2023, Google released another AIGC music model, MusicLM. Compared to AudioLM, it marked a series of sound clips and mapped them to semantic segments for training. This means that this model can receive user text or audio input and generate AI music works. However, Google stated that, considering the "potential protests from musicians and copyright risks," it has not formally released this model and only conducted small-scale open testing.
Undoubtedly, from MusicLM to Lyria, with more training on music materials, the music generated by the Lyria model is more accurate and profound. However, more importantly, AIGC music models are being accepted by more musicians, and Google is gradually learning to avoid the copyright risks of AIGC in its game with record companies.
How to Establish Business Rules for the AIGC Music Era?
Let's go back 7 months.
An AIGC song called "Heart on My Sleeve" went viral across the internet. Soon, this song angered the record companies behind Drake and The Weeknd, Universal Music. Under Universal Music's complaint, "Heart on My Sleeve" was taken down from the entire internet.
Now, the long-tail effect brought about by this song has prompted YouTube to take the restrictions on the use of AIGC music seriously.
In a public blog post on November 14, YouTube stated that it "will introduce a new right for music partners on the platform," allowing them to request the removal of AIGC content that "mimics the unique singing or rapping of artists." The platform will consider many factors when evaluating these requests, such as whether the music comes from "news reports, analysis, or commentary."
At the same time, in August of this year, YouTube reached a partnership with Universal Music to jointly develop AI music tools. Therefore, these controversial contents will also be open to the music creators and record companies of the AI tool development team, and in the coming months, YouTube will continue to expand the scope of openness.
It is evident that whether the materials used to train AI models comply with copyright regulations is the core of the dispute between music platforms and record companies. Avoiding copyright risks and gaining the support of artists are the common goals of all parties.
Following this logic, Spotify chose to use AI to "clone" celebrities in the podcast field, rather than in the more controversial music field.
In September of this year, Spotify launched a new AI voice cloning tool. With the technical support of OpenAI's automatic speech recognition model Whisper, individuals can automatically switch to various languages such as Spanish, French, and German, and it is "completely authentic," even reproducing the rhythm and intonation. This not only did not provoke dissatisfaction in the podcast market, but also received authorization from many famous podcast hosts such as Dax Shepard, Monica Padman, and Lex Fridman.
In addition to settling disputes over music content in terms of copyright, platform providers are also strengthening the identification and notification of AIGC content.
It is reported that YouTube's new rules will require creators to disclose their "synthesized or modified content" and indicate the "materials modified or synthesized," as well as the "AI tools used." YouTube will add a new label to the description page, indicating that the content's "sound or image has been altered or synthesized."
This move by YouTube may have been inspired by TikTok. As early as September 19 of this year, TikTok required creators to label content created with AI as "AI-generated." In addition, TikTok also stated that it is testing new automated tools to label "content we detect as edited or created with AI," which has become a focus of many platforms' regulation of AIGC.
In a public blog post, YouTube stated that it will focus on deploying "machine learning technology combined with human reviewers" to "enforce community guidelines." "AI tools help detect potential violations on a large scale, while reviewers work to confirm whether the content has indeed crossed policy boundaries," which may refer to SynthID, expanded to the music field for the first time.
At the same time, Believe also announced AI music recognition technology and publicly disclosed its accuracy.
Believe recently announced in its Q3 financial report that they have "created some exciting AI technology" that can determine with 98% accuracy whether music is AI-generated and with 93% accuracy whether it is a "deepfake." The announcement stated that this technology, known as "AI Radar," is gradually being applied.
From copyright protection for AIGC music, user notifications, to related recognition technology, platforms such as Google, Spotify, Believe, and TikTok have taken roughly similar regulatory paths. However, facing the proliferation of AIGC, on the other side of the coin, platforms are also intensively accelerating their layout in the AIGC field.
As mentioned earlier, YouTube has partnered with Universal Music to launch a music AI incubator, collaborating with artists under Universal to explore the application of artificial intelligence in music. Artists can get early access to developing AI products and provide feedback. The Music AI Tools fully consider the opinions of these artists.
At the same time, YouTube is also in talks with the three major record companies, hoping to use their copyrighted music for AI model training. However, from the current Dream Track collaboration artists, who are mainly concentrated in the music industry under Universal Music, Warner Music, and other record companies, it seems that Sony Music has not yet been included.
When the opportunities for creating AIGC content in collaboration with record companies are very limited, platform providers are finding alternative paths.
For example, in December 2022, Deezer launched the Zen app, mainly providing users with "health guidance, exercise audio, and immersive sleep and healing music and audio." Currently, Deezer is using AI technology to produce content for the platform.
In China, NetEase Cloud Music also launched the AI music production platform Xstudio in collaboration with Xiaoice Company, providing AI sound sources for music creators. TME's Galaxy Sound Effects, MUSE, Tianqin Lab, and Tencent AI Lab have jointly created the intelligent music creation assistant TME Studio, integrating a series of AI functions such as music separation, MIR calculation, lyric assistance, and intelligent scores.
It is obvious that in the wave of AIGC music, record companies and musicians holding the rights have always been at the forefront of technological limitations. And all platforms are making efforts to coordinate, striking a balance between allowing innovative behaviors on the platform and protecting works that are protected by copyright.
As Deezer CEO Jeronimo Folgueira said, "For streaming platforms, AI is a big problem, but it's an even bigger opportunity."
The AIGC wave will not recede, and the UGC era has already provided the answer
At the end of September 2023, Robert Kyncl, the new CEO of Warner Music Group, attended the 2023 Code Conference in California. In his speech, he pointed out that the analogy between AIGC and UGC is very clear. The threat of AI technology can also find solutions from UGC.
Before taking office at Warner Music, Kyncl worked at YouTube and Netflix for 20 years, helping YouTube pioneer the era of UGC content and YouTube influencers, maintaining its competitiveness globally. This experience has made Kyncl less angry and more peaceful when facing AIGC, after all, AIGC "will not recede."
In his speech, Kyncl reviewed YouTube when UGC was just emerging. At that time, YouTube often received complaints from a large number of content owners, stating that their videos or music were uploaded to the platform without their permission.
It wasn't until 2006, after Google acquired YouTube, that a copyright recognition technology called Content ID was established to track copyrighted content on the platform and give the corresponding rights to copyright owners, allowing them to remove or benefit from this content.
"This way, we have built a business worth billions of dollars, which is now a business worth billions of dollars every year," Kyncl said. "This is an incredible new source of income for everyone. And AI is such a new super tool. So, we need to treat it with the same mindset."
Following YouTube's governance logic in the UGC era, Synth ID technology seems to have the potential to become the next existence that ensures the interests of all parties in the AIGC wave, and the next step is negotiating the allocation rules between platform providers and copyright holders.
In other words, when the platforms mark this AIGC music, how will they handle this music? How will they distribute the copyright revenue among the different entities involved? What proportion of the revenue will be shared among the music used to train the AI corpus? This will be the key to whether the platforms can establish sustainable business cooperation relationships.
However, although various issues need to be resolved, the direction that platform providers are striving for is clear and definite.
From PGC to UGC and now to AIGC, each change in the discourse power of internet content production faces a new round of questioning and regulation. Just as in the past, to meet the needs of the development of the UGC wave, short video platforms entered the era of centralized purchase of music copyrights.
Today, the copyright issues faced by AIGC are even more complex. But in the face of the outbreak of productivity and the enthusiasm for content consumption, platform providers will eventually walk the old path of the UGC era again, engaging in a long-term copyright tug-of-war with creators and copyright owners.
However, whether it is the platform providers or the record companies, everyone's direction is actually the same, just exploring operational rules on the road forward.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
