Cloning Voices

New voice apps are making it increasingly possible to imitate a person’s speech easily and precisely. Originally developed as an assistive technology for people at risk of losing their ability to talk (due to cancer or surgery, for example), the ability for virtually anyone to clone speech patterns opens up a host of issues from harmless pranks to serious security breaches.

The apps are available for a variety of languages, including French, English and four widely spoken Indian languages. To develop a simulated voice library, these apps require anywhere from 50 to a few hundred phrases/sentences, and they work using even decent-quality recordings, such as YouTube videos. The apps store snippets of speech as short as five milliseconds long that can be shuffled together to create words. And developers are already planning to enhance the apps to add emotive qualities (like happiness and sadness) to individual words or phrases.

Although there are many constructive applications for this new technology, it will likely render current voice security systems obsolete. Tests performed using one of the available apps demonstrated that it was able to fool a bank voice biometric system 80 percent of the time. And humans (with supposedly better abilities to discern subtle sound differences) were only able to detect the cloned voice with a reliability rate of 50 percent.