Compile | Pang Xiaochun
Edit | Cheng Qian
Zhidong News on May 11, yesterday, ElevenLabs, a voice artificial intelligence startup, released an early preview of a music generation artificial intelligence model, which can generate song samples with lyrics according to text prompts.
ElevenLabs is led by Piotr Dabkowski, a former machine learning engineer at Google, and Mati Staniszewski, a former strategist at Palantir, an analytical software company, whose large models of music can be used in a variety of scenarios. However, since its release, the model has not been well received by everyone, and many people have expressed concerns about the originality and use of the data it uses.
1. The publicity strategy imitates Sora, and collects netizens to generate music creatively in real time
ElevenLabs specializes in using machine learning (ML) for speech cloning and synthesis in different languages, and the company has launched a variety of speech generation capabilities, including text-to-speech, voice changers, voice-overs, and more. Unsurprisingly, the company has set its sights on the music industry.
To raise awareness of the model, ElevenLabs used Sam Altman's approach to the launch of Sora, a video model, to solicit ideas on social media and convert texts provided by netizens into music.
It's unclear when the model can generate music, but based on an example posted on X by the company's head of design, Ammaar Reshi, VentureBeat guesses that the model will be able to generate lyrics for a three-minute piece of music.
VentureBeat cites some of the use cases for music mockups: generating a fun lullaby to lull kids to sleep, creating a clever advertising song for a marketing campaign, and designing a lively musical intro for a podcast show. Foreign media believe that some people may use this model to develop the next hit song, and said that a large number of music artificial intelligence startups have emerged, including Harmonai, Lyrical Labs, Suno AI, Loudly, etc.
ElevenLabs launched a product in January called Voice Library, which allows users to sell these AI-generated songs through the voice library. Currently, the company's voice library already allows users to sell their own AI clones of voices, and the company controls the range of voices available to them.
Second, the copyright of the training data is questioned, and it may also be used for deepfakes
However, music generation technology is not popular with everyone. As with all generative AI applications, the question for ElevenLabs is what kind of data this model is built on, and whether the training data contains copyrighted content. If so, whether ElevenLabs has permission from the owner of the content, or whether the content is reasonably protected.
There are also those who oppose music generation technology who argue that artists may lose their jobs. AI will easily replicate an artist's style and then launch new music on their behalf. It is also possible that such techniques could be used to produce deepfakes. VentureBeat has reached out to ElevenLabs for more comments on its music generation model, and said it would update the media coverage if it receives a response.
Conclusion: The model is not yet mature, and there are both opportunities and challenges
ElevenLabs' music generation model is still in early preview, but it has already shown great potential and influence. It can quickly produce music works and improve the efficiency of music production, but it may also have a certain impact on traditional musicians.
The emergence of music generation models heralds a possible technology-driven change in the music industry, bringing both opportunities and challenges.
来源：VentureBeat