OpenAI Clarifies Text-to-Speech Tool Status and Addresses Concerns
TapTechNews reports on June 10 that OpenAI has for the second time in a few months explained about its text-to-speech tool and emphasized again that the tool is not yet widely available and may not be so in the future.
Whether we eventually deploy this technology on a large scale or not, it is extremely important to let people around the world understand the direction of this technology's development, OpenAI said in a statement posted on its website on Friday. That's why we want to explain how the model works, how we use it for research and education, and how we implement safety measures around this technology.
According to TapTechNews' understanding, at the end of last year, OpenAI shared its voice engine with a small number of external users. The engine can generate a natural-sounding human voice that closely resembles the original speaker by using text input and a 15-second audio clip of a human voice. This tool can create multilingual character voices that are indistinguishable from the real ones. At that time, the company said that they chose to preview this technology rather than release it on a large scale to enhance society's resilience in response to the threat posed by increasingly realistic AI-generated models.
As part of these efforts, OpenAI said they are actively phasing out the use of voice recognition for bank account verification, exploring policies to protect the use of individuals' voices in AI, educating the public about the risks of AI, and accelerating the development of technologies to track audio-visual content so that users can identify whether they are interacting with a real person or a synthesized content.
However, despite these efforts, concerns about this technology still exist. Bruce Reed, the head of artificial intelligence for US President Joe Biden, has said that voice cloning technology is one of the reasons he loses sleep at night. The US Federal Trade Commission (FTC) said in March that scammers are using AI techniques to enhance the credibility of their fraudulent activities. They use voice cloning tools to make it harder for people to distinguish between AI-generated voices and human voices.
In its updated statement on Friday, OpenAI tried to alleviate these concerns. We continue to engage with US and international partners from government, media, entertainment, education, civil society, and other fields to ensure that we incorporate their feedback in the construction process. The company also noted that once the voice engine is equipped with its latest model, GPT-4o, it will also bring new threats. The company said that internally they are actively conducting'red team testing' on GPT-4o to identify and address known and unknown risks from areas such as social psychology, bias and fairness, and misinformation.