OpenAI has unruled its text-to-speech model Voice Engine. OpenAI claims that the model can clone the voice of any person by taking 15 seconds of the audio clip. The company has now shared it with about 10 developers and plans a wider rollout.
This model can be used to generate audio or a person, translate content in the same voice, and more. The company developed Voice Engine back in 2022 and has already been using it in the text-to-speech API as well as ChatGPT Voice and Read Aloud. But it didn’t do a wider rollout to avoid misuse of synthetic voice.
“We recognize that generating speech that resembles people’s voices has serious risks, which are especially top of mind in an election year,” the company wrote in a blog post
OpenAI says that this model can be helpful in many ways. It can be used to translate a video into multiple languages in the speaker’s voice. Creators who publish Podcasts can use this model to reach a wider audience globally. Teachers can use it to provide interactive voice assistance to their students without needing to respond to them manually.
During the testing phase, OpenAI requires developers to agree to its usage policies. The company wants developers to get consent from the original speaker to use his voice and also disclose to listeners that the voice is AI-generated. The company also adds an inaudible audio watermark to distinguish if the audio was created using OpenAI’s Voice Engine.
OpenAI says that it will scale the model after receiving feedback from stakeholders such as policymakers, industry experts, educators, and more. It doesn’t want the misuse of this tech.
