Nvidia Unveiled AI Model for Transforming and Creating Audio

 

On November 25, Nvidia (NASDAQ: NVDA) introduced a groundbreaking AI model, Fugatto, designed to generate original sounds and modify existing audio. Its target customers are creators in music, film, and gaming. The leading AI chips and software company has not planned to put Fugatto on the market for customers yet.

 

Fugatto, short for Foundational Generative Audio Transformer Opus 1, can create sound effects from text descriptions, including innovative sounds like making a trumpet bark like a dog. Beyond generating new audio, the model’s standout feature is its ability to transform existing audio. For instance, it can convert instrumental sounds into human sounds or alter spoken recordings by changing accents or moods. These peculiarities set it apart from its constituent technology, Meta (NASDAQ: META)'s conversion of text to audio.

 

Bryan Catanzaro, Nvidia’s Vice President of Applied Deep Learning Research, emphasized that generative AI sound could bring new ways to create music, video games, and other entertainment. However, due to audio copyrights, it is controversial whether AI voices can be applied to entertainment works.

 

Fugatto was trained on open-source data, and Nvidia is deliberating its release strategy to avoid risks of misuse such as generating misinformation or violating copyrights. Catanzaro said that Nvidia needed to reduce the risks of the AI model before putting it on the market. That is why Nvidia does not release it publicly.

 

As of now, NVIDIA has not given a plan to prevent AI models from being misused. Meta and Open AI, which develop audio-making AI models like Nvidia, have also not made their AI models public.