OpenAI has introduced a new voice cloning tool, Voice Engine, which is currently in testing and allows cloning voices from a 15-second audio sample. However, the date of the tool’s public release remains uncertain, as OpenAI is still assessing the potential misuse of this technology.

The voice engine, which has been in development for about two years, uses a generative AI model that is also built into ChatGPT’s voice capabilities and OpenAI’s text-to-speech API. Spotify has been using this model since early September to dub podcasts in different languages. Its training data, which is a combination of licensed and publicly available sources, remains a closely guarded secret due to competitive and legal considerations.

Voice sample:

Voice generated from a sample:

So far, OpenAI has taken steps to ensure the ethical use of the Voice Engine, including marking cloned voices with special signs with identifiers inaudible to the human ear and limiting initial access to a select group of developers who focus on low-risk, socially beneficial use cases. The company also requires users to obtain explicit consent from individuals whose voices are cloned and to disclose when voices are generated using artificial intelligence.

Voice sample in English:

German translation has been generated:

However, the emergence of advanced voice cloning technology poses potential threats to voice actors’ livelihoods and raises concerns about the creation of convincing fakes that could be used for fraudulent schemes.

OpenAI’s roadmap for the Voice Engine envisions further security enhancements and potentially broader access, depending on the results of the pilot program and the development of effective protections.