Microsoft Decides It’s Too Dangerous to Launch Voice-Cloning AI

Microsoft’s AI tool that can clone people’s voices is too dangerous to be released to the public, say researchers involved in the project. The model will now be used for research only..

The model, known as VALL-E 2, is a neural language coder, a type of artificial intelligence that uses neural network techniques to encode and decode linguistic information.

“We currently have no plans to include VALL-E 2 in a product or expand its access to the public,” Microsoft said on its official website. “This could lead to potential risks in misuse of the model, such as falsifying voice identity or impersonating a specific speaker,” it added.

Unlike the original VALL-E, which was launched in January 2023, VALL-E 2 can synthesize sounds not presented during trainingsynthesizing sounds based on a short audio clip, copying the intonation and various details of the original sound.

However Microsoft recognizes that the model may be useful in other contextssuch as education, entertainment, journalism, authored content or accessibility.

However, the company notes that the model is not yet perfect. “Although VALL-E 2 can speak with a voice similar to vocal talent, the similarity and naturalness depend on the length and quality of the speech prompt, background noise and other factors,” the company emphasized.

VALL-E 2 is not the only model in the category

While Microsoft is aware of the risks of releasing a tool like VALL-E 2 to the public, other companies are testing similar AIs. One example is OpenAI, which introduced its Voice Engine in April. Model that can copy someone’s voice in just 15 seconds.

Similarly, OpenAI is aware that the technology can be used for educational purposes or to provide accessibility. The company sees the tool as a solution for people with disorders that impede or prevent speech.

There are currently commercial solutions exploring the use of AI to replicate the voices of real people. For example, Truecaller’s AI Assistant uses Microsoft Azure AI Speech to answer calls by voice.