ElevenLabs, a company that makes AI audio tools, released SFX v3 on Oct. 29, a new version of its AI that turns a written description into a sound effect. The company says it now does better when one description asks for several sounds. Its example is a football clip that needs a crowd cheering, a ball dropping and the same crowd booing seconds later, which until now took trial and error to get from a single request.
ElevenLabs says more of the sounds a person describes now make it into the result. It pitches this at people making videos and games: describe the main action, add the sounds around it, listen to how they work together, then change the words to get closer to what you had in mind.
The new version also does better when an effect includes music or voices, such as a gospel choir singing hello, the company says. And its sounds should come out cleaner: ElevenLabs says gongs, chimes and whistles are less likely to turn harsh or crackly, so they better keep the ring or tone you asked for.
SFX v3 is available in ElevenCreative, the company's app for making audio and video, and in its phone app, and developers can build it into their own products. But ElevenLabs' guide for developers says access is limited and may not reach every region, and the older version stays the default for them.
So far the only word on these improvements comes from ElevenLabs itself. Its guide says results vary from one description to the next and suggests comparing the old and new versions on your own requests. It also says the AI doesn't always follow a request for an exact number of repeats, such as a triangle struck four times. The announcement does not say what sounds the AI learned from.