OpenAI Launches New Voice Cloning Technology: Replicating Your Voice in Just 15 Seconds
-
According to media reports, OpenAI has recently introduced a revolutionary voice cloning technology 'Voice Engine'.
Voice Engine can generate highly realistic, emotionally rich, and natural-sounding speech that closely resembles the original speaker's voice with just text input and a 15-second audio sample. The development of this technology began in 2022 and has already been implemented in the company's existing text-to-speech APIs and preset voices in the Read Aloud feature.
OpenAI believes that Voice Engine technology holds significant importance for multiple fields. In reading assistance and language translation, it can provide more natural speech output, enhancing user experience.
At the same time, this technology is a major boon for individuals with speech impairments, helping them communicate more smoothly. For example, in a pilot project at Brown University, the technology was successfully used to create voice clones extracted from audio recordings of school projects, effectively assisting students with speech impairments. However, given the potential misuse risks of synthetic voice technology, OpenAI is currently only conducting small-scale testing with a limited number of trusted partners. Through this approach, the company aims to gain deeper insights into the technology's potential applications and evaluate possible risks.
OpenAI also hopes this initiative will spark broader societal discussions on the responsible deployment of synthetic voice technology, collectively exploring how to adapt to this emerging technology.
Additionally, to ensure the safe use of the technology, OpenAI has implemented a series of safety measures. These measures include using watermarking technology to track audio sources and actively monitoring system usage patterns. When the product is officially launched to the market, the company will establish a 'voice blacklist' to detect and block AI-generated voices that are too similar to those of celebrities, thereby avoiding potential copyright and privacy issues.