AI Audio Generators and ‘Her’

In 2021, SBS’s New Year Special Program 〈Fight of the Century: AI vs human〉 covered a composing competition between South Korea’s first AI songwriter EVOM, and human songwriter Do-il Kim.

Image from Google Deepmind’s V2A Technology Introduction Page

Three years have passed, and AI audio generators have made rapid growth. Stability.ai and ElevenLabs announced their technology could generate a maximum of three-minute-length music and sound effects, including “People cheering in the baseball stadium” with text prompts. Also, Google DeepMind recently revealed a V2A(Video-to-Audio) technology that could ‘understand’ the videos users upload and generate soundtracks that match the atmosphere.

AI audio generators brought a new wave to music by enabling people to compose through text. However, when the ‘human voice’ intervenes, it guides us to the other direction we should not overlook. For example, IU’s ‘AI cover’ of the song 〈밤양갱(Chestnut Paste Jelly)〉 made with machine-learned IU’s voice, does not serve IU’s neighboring copyright and her personality right due to its process.

Furthermore, this case stretches to the case of ‘imitating’. Open AI’s latest multi-modal, GPT-4o received handsome praise for its ability to recognize voice · text · video, yet was criticized for violating Scarlett Johansson’s publicity right who took the role of AI assistant in the movie 〈Her〉 by ‘impersonating’ her voice for their voice module, ‘Sky’.

This escalated particularly after it was revealed that Johansson turned down Open AI’s proposal to use her voice for Sky’s beforehand, and Open AI’s CEO Sam Altman GPT-4o used the word ‘her’ to promote GPT-4o on his social media.

Overall, the AI audio generators’ technology and usage will not be stopped at the aspect of the methodology. Its future will be formed with the discourse surrounding ethics, regulations, and limitations.

References