The Chinese company Baidu has developed an artificial intelligence with scary powers: it can mimic any voice after hearing a person speaking for less than a minute.
This is not the first search that tries to do this. In the past, a company developed a neural network that could replicate the voice of US President Donald Trump and former President Barack Obama from some of their speech samples.
The idea of Baidu, however, is not to contribute to the dissemination of false news. “We are going to have many possible applications for this technology,” said Leo Zou of Baidu’s communications team, Digital Trends. “Voice cloning can help patients who have lost their voices,” he explained. “A mother can easily set up an audiobook reader with her own voice,” he said, recalling the possibility of using technology in human-machine interfaces.
To develop the synthetic voice system, Baidu researchers used more than 800 hours of audio from 2,400 different people. With some stretches of audio, they were able to accurately reproduce the voice more than 95% of the time.
The study was published in arXiv and can be read in full here.