Baidu AI can mimic voices after 1 minute of training

There are concerns the technology could be misused

The Chinese search engine Baidu has created an AI system that can mimic a person's speech after listening to just 50 seconds of audio. The system is capable of “speaking” with the person's voice after being fed ten 5-second snippets of the person's speech. The resulting voice is convincing enough that it is able to fool a voice recognition system over 95% of the time. Some commentators are concerned that the technology could be misused by fraudsters, who could use it to trick voice-based authentication systems and gain access to private information. Baidu itself is keen to stress that the AI has much more benevolent applications, such as a parent configuring an audiobook to read to their child in their voice. You can listen to audio samples of the AI at https://audiodemos.github.io/ 

Individual Basic

Free
 
  • 1 Knowledgebase
Subscribe free

Individual Premium

£15 Per month
Cancel anytime
  • Unlimited
    Plus 50.00% off International Search Summit ticket
Subscribe

Team global

£95 Per month
Up to 7 users, - additional users £10 per month - cancel anytime
  • Unlimited
    Plus 2 tickets for International Search Summit
Subscribe