快猫短视频s who have built a computer program capable of mimicking human voices are also developing watermarking technology to prevent the system from being misused.
The program, developed by a team at AT&T Laboratories in the US, convincingly mimics a particular voice by reconstructing its individual nuances and intonations from pre-recordings. The technology is far from perfect but its creators are wary that it could be misused in the future. The system is most likely to find its way into call centres, text-to-speech software and automated newswire reports.
Juergen Schroeter, a speech synthesis expert at AT&T, told 快猫短视频 that the technology has plenty of commercial potential, but also admitted that he has security concerns.
Advertisement
鈥淲hat happens later, if someone steals a voice, is a problem,鈥� he says. 鈥淭hat鈥檚 where watermarking comes in.鈥�
Speech synthesis
There are still traces of unnatural, computerised tones, in the voices that have so far been generated by AT&T. Creating each new synthesized voice requires between 10 and 40 hours of studio recordings of the person speaking and a powerful computer to process all the data. Nevertheless, the experts who built the program believe that within a few years, it could create a perfect copy of someone鈥檚 voice.
Schroeter says it will be possible to add information to an audio signal that would not be detected by the human ear but could be picked up by a special detector. This would enable another computer to detect when a voice has been generated artificially by this system, and when it is genuine.
Schroeter and fellow researchers are already designing such a system, but he says that voice mimicking technology is not yet advanced enough to require it. 鈥淲e鈥檒l look at that when we鈥檙e good enough,鈥� he says.
Schroeter鈥檚 vision is to one day build a system that would allow a person to make a copy of their own voice within minutes. 鈥淵ou could imagine a company chief executive donating their voice for the help desk,鈥� he says.
AT&T has spun off a company, called Natural Voices, to market the voice copying system.