Baidu, the equivalent of Google in China has released a white paper that shows its latest development in AI. It is a program that can clone voices even after a seconds-long clip with the help of neural networks. The software is not only able to clone voices inputted to the device but can change them to reflect another gender and a different accent.
You can listen to some of these generated samples on GitHub. This is not something new. Software has been able to clone voices in the past as well but it took much longer. What previously took half an hour has now been converted into a few seconds.
There are some useful applications for this technology. Imagine you are far away on a business trip and your kids are at home. They can be read a bedtime story in your voice. A person who has lost the ability to talk can have a duplicate voice created. This tech will also greatly enhance the digital assistants and will offer more personalized services.
However, as is the case with all technology, there are good applications coupled together with the bad. According to New Scientist, the program was able to produce a voice that was able to fool voice recognition software with a 95% accuracy. The cloned voice was rated 3.16 out of 4 by humans. This can be used for good as well but it can also be used for AI-assisted fraud.
People are already easy to fool with fake written news and photoshop. It is not hard to imagine the chaos that could ensue in the case of this technology falling into wrong hands. You can watch the video below to see how it can impact us:
