OpenAI’s Voice Engine is a publicly available program that enables users to create voice imitations based on a short 15-second speech excerpt. The outcomes are remarkably precise, allowing users to produce voice replicas that closely resemble the original speakers using the Tone Clone feature. Moreover, users have the ability to generate imitations in various languages, aligning with the tool’s primary objective. AI technology has significantly enhanced the field of robot speech synthesis. Well-known robotic voices such as Siri, Alexa, and the unique TikTok voice exhibit distinct pauses, intonations, and inflections that set them apart from conventional human speech. The scripted phrase “When you are finished recording, you can hang up or press pound. For more options.” showcases a robotic delivery style that is distinguishable from natural human speech patterns.
Nevertheless, differentiating between a sound file engineered to emulate a real person and an actual individual delivering the same content can be challenging. Elements like tone, pauses, and voice nuances all contribute to the distinctiveness of a person’s speech patterns. Upon careful scrutiny of these 15-second simulated clips, one may discern subtle deviations from natural speech cadence. Nonetheless, these clips bear a striking resemblance to the expressive voice of a skilled voice actor.
The implications for the voice acting industry remain uncertain. In scenarios requiring professional delivery, such as a car insurance advertisement, consistency and vocal quality play pivotal roles. While AI-generated voices can effectively achieve this, it may still be simpler to request a human performer to emphasize specific words or alter the emotional undertone of the delivery. Directing a system to sound “happy” often yields a cheerful, TikTok-like demeanor. Conversely, Voice Engine’s narrative progression could evolve from restrained curiosity to heightened excitement and joy. These advancements may eventually elevate AI systems to the level of trained voice actors, sparking legal and ethical debates that are currently emerging.
As the US government takes measures to curb the misuse of AI voice technology, the Federal Communications Commission has prohibited the use of AI-generated voices in robocalls following instances of emails featuring an AI-cloned voice of President Joe Biden.
OpenAI has established stringent guidelines for partners utilizing its Voice Generation technology. These guidelines mandate that partners refrain from using the technology to deceive individuals or entities without explicit consent. Partners must secure the original speaker’s explicit and informed approval, prevent unauthorized voice replication, and transparently disclose that the voices are AI-generated. OpenAI has implemented watermarking on the sound clips to trace their origins and monitor their usage continuously.
In addition to discontinuing voice-based authentication for bank account access, OpenAI has proposed a series of measures to mitigate the associated risks, including safeguarding individuals’ voices in AI applications, enhancing awareness of AI deepfakes, and developing tracking mechanisms for AI-generated content.
Emilia David, The Verge
