OpenAI demos voice cloning tech but delays launch over risks.
AI startup OpenAI has showcased new technology capable of mimicking human voices from short audio clips. However, it has decided against a public rollout citing potential impersonation dangers. The company founded by Sam Altman unveiled its ‘Voice Engine’ that can recreate speech with just 15 seconds of recording. But OpenAI said early testers are being advised against impersonating others without consent.
Authorities in the US are probing robocalls using AI-modified voices ahead of New Hampshire polls that featured what seemed like Biden’s voice. Such incidents underscore the possibility of misuse.
Several firms currently sell voice mimicry programs, though OpenAI fears its superior system could amplify risks. It had adopted a similar approach for its video generator ‘Sora’.
Meanwhile, the company filed trademarks for voice and speech tech, indicating plans to enter that market segment and take on leaders like Amazon’s Alexa. But safety remains a work in progress.
While improving the nascent technology could help OpenAI compete, delayed launch reflects the sensitivity around voice deepfakes, especially in election seasons. Its regulated preview testing is aimed at responsible development.