OpenAI unveiled its most recent Voice Engine model, which can produce speech that sounds natural given text input and a short audio sample of 15 seconds.

The company stated that although the technology can replicate a human voice from a 15-second recording, it would not be made available to the public until it has further information on any abuse hazards.

Notably, Voice Engine can use very little audio input to produce voices that are genuine and emotional.

According to OpenAI, the Voice Engine project started in late 2022 and was initially used to power the read-aloud, ChatGPT Voice, and preset voices in OpenAI's text-to-speech API. Like its text-to-video creation model, the corporation has not yet made it available to the public due to worries about possible misuse.

Voice Engine has went through private testing by OpenAI with a select number of reliable partners. Promising applications have been found in a number of fields after early studies. Among them are:

Improving Education: Voice Engine is used by education technology business Age of Learning to provide voice-over content that is pre-written and intended to help youngsters and non-readers with reading.

Global Content Translation: Voice Engine is used by HeyGen, an AI visual storytelling platform, to translate podcasts and videos into a variety of languages while maintaining the original speaker's natural accent.

Community Health Services: Dimagi improves the delivery of critical services in remote locations, especially in healthcare settings, by utilizing Voice Engine and GPT-4.

Assistive Communication: Voice Engine is integrated into Livox, an AI communication software, to provide non-robotic, customizable voices for those with speech-related disabilities.

Clinical Applications: Voice Engine's ability to restore speech for patients with medical conditions-related speech impairments is being investigated in clinical settings by Lifespan's Norman Prince Neurosciences Institute.