OpenAI built a voice cloning tool, but you can’t use it… yet

OpenAI's Voice Engine is a new addition to the company's text-to-speech API, which allows users to upload a 15-second voice sample and generate a synthetic copy of that voice. It's an AI-powered tool that has been under development for approximately two years and is currently in preview debut, with no public availability date announced yet.

To use OpenAI's Voice Engine, you need to upload any 15-second voice sample, and the system will generate a synthetic copy of that voice. The technology analyzes both the speech data from the sample and the text data meant to be read aloud, creating a matching voice without the need for a custom model per speaker.

The ethical concerns with OpenAI's Voice Engine include:

• The possibility of voice cloning being used to create deepfakes that could threaten the livelihoods of voice actors.
• The risk of the technology being abused to spread misinformation or for malicious purposes, such as generating hateful messages or fooling authentication systems.
• Concerns over the use of copyrighted material in training AI models and the lack of an opt-out scheme for voice data.

OpenAI's Voice Engine is listed as costing $15 per one million characters, which translates to around 18 hours of audio. This pricing is considered aggressive and cheaper than some rival vendors, such as ElevenLabs, which charges $11 for 100,000 characters per month. However, Voice Engine does not currently offer controls for fine-tuning the tone, pitch, or cadence of a voice.

To prevent misuse of Voice Engine, OpenAI is implementing several measures:

• Limiting access to a small group of developers focusing on low-risk and socially beneficial use cases.
• Watermarking clones with inaudible identifiers to track and identify generated clips.
• Providing access to its red teaming network to identify potential malicious uses.
• Testing a security mechanism that requires users to read randomly generated text to prove they're aware of how their voice is being used.