OpenAI Updates Voice Engine: Human-Like Audio from Text Using 15-Second Sample Amid Safety Concerns

Jun 8, 2024, 06:46 PM

OpenAI has published an update on its Voice Engine, a text-to-speech (TTS) model capable of generating human-like audio from text using a 15-second sample of speech. The Voice Engine employs a diffusion process to match the speaker's voice, starting with random noise. This technology, which was demonstrated to global policymakers last summer, has raised concerns about safety and potential misuse, particularly related to the Sky voice. OpenAI's existing collection of TTS voices, built using professional voice actors, also uses a 15-second sample to define each voice. The company has been addressing these concerns through ongoing safety research. OpenAI had a voice engine in 2022 and has been transparent about its development.