According to leaked reports, OpenAI is working on a tool capable of creating music based on text and audio prompts. This tool could be used, for example, to add music to existing videos or a guitar accompaniment to an already recorded vocal track. Sources close to the development suggest that OpenAI is collaborating with students from The Juilliard School in the US, who are annotating musical scores to create a database for training the model. However, a representative of The Juilliard School denied these reports, and OpenAI did not respond to inquiries.
It is not yet clear when OpenAI plans to launch the tool or whether it will be available as a standalone product or integrated into existing applications such as ChatGPT or the Sora video tool.
OpenAI's Past Experiments with Music AI
OpenAI has experimented with generative music in the past. In 2019, it launched MuseNet, a neural network capable of creating four-minute compositions featuring ten different instruments. A demonstration by the company showed how MuseNet could rework Wolfgang Mozart's Rondo Alla Turca in the style of Frederic Chopin.
A year later, in 2020, Jukebox followed, another neural network that generates music, including basic vocals in various genres, based on data about artists and lyrics. These earlier models predated the major success of ChatGPT but indicate OpenAI's long-term interest in music applications.
Competition and Collaboration in the Industry
OpenAI's move comes at a time when other companies are also developing similar technologies. For example, Spotify recently announced a partnership with major music groups such as Sony, Universal, and Warner to develop responsible AI products. Spotify already uses AI to create personalized playlists, such as "daylist," and its AI DJ feature, which suggests related songs based on tracks users have listened to. The platform also hosts some AI-generated music.
Startups in the field include Suno and ElevenLabs, which have their own platforms for AI-generated music. Another player is Google, with its generative music models. OpenAI could thus enter into direct competition with Suno, potentially affecting the AI music market.
Copyright and Controversy
The development of AI music tools is raising copyright concerns. European music groups such as the European Composer and Songwriter Alliance (ECSA) and the European Grouping of Societies of Authors and Composers (GESAC) previously warned that the 2024 EU AI Act does not adequately protect creators whose works are used to train generative AI models. Without a clear way to opt out or receive payment, artists' work remains under constant threat.
Suno is facing a lawsuit in the US from the Recording Industry Association of America (RIAA), which claims that the company pirated songs from YouTube to train its models. Universal Music Group is also pursuing a copyright lawsuit against the AI company Anthropic. These cases highlight the risks associated with using existing works to train AI.
Sources: techcrunch.com and euronews.com
