快猫短视频

Chinese search giant Baidu says it can create a copy of someone鈥檚 voice using neural networks 鈥� and all that鈥檚 needed to work from is less than a minute鈥檚 worth of audio of the person talking.

Baidu researchers say the technology could create digital duplicate voices for people who have lost the ability to talk. It could also be used to personalise digital assistants, video game characters or automatic speech translation services.

鈥淎 mum could easily configure an audio-book reader with her own voice to read bedtime stories for her kids,鈥� says Sercan Arik at Baidu Research, who led the work.

Voice cloning technology has improved rapidly in recent years. In 2016 Adobe released VoCo, which could mimic someone鈥檚 voice using 20 minutes of audio. Last year, released a service letting anyone create a digital copy of their voice based on 1 minute of audio.

Digital mimic

Baidu鈥檚 research builds on its text-to speech synthesis system Deep Voice, which was trained on more than 800 hours of audio from 2400 speakers. It builds a model of human speech by learning what sounds go with what text and also learns the idiosyncrasies of each speaker it was trained on.

Now new software is able to synthesise a copy of a voice based on just hearing snatches of the original. The best version needed 100 snippets, each no more than 5 seconds long, the Baidu team says. But one trained on just 10 snippets performed well enough to dupe a voice recognition system more than 95 per cent of the time, and human evaluators gave it 3.16 out of 4 for mimickry.

The output is still not totally indistinguishable from the human voice, says Arik, 鈥渂ut it does show a very fundamental breakthrough in that direction鈥�. You can listen to the voices

Rita Singh, a voice forensic science expert at Carnegie Mellon University, points out that even the best synthesised voices contain telltale digital signals easily detected by advanced voice profiling algorithms.

But most voice authentication systems 鈥� used to secure everything from banking services to smartphones 鈥� can be fooled because they rely instead on picking up broad statistical features, she says.

In 2014, Nitesh Saxena, a security researcher at the University of Alabama at Birmingham, showed that a freely available voice morphing tool could trick voice authentication systems 80 to 90 per cent of the time. Unpublished research shows leading digital assistants and even a major bank鈥檚 telephone service remain vulnerable, he says.

But while biometric systems can be improved, our own ability to detect fakes can鈥檛. This raises the spectre of voice synthesis systems aping someone鈥檚 voice to commit fraud or sparking fake news by doctoring a politician鈥檚 speech.

鈥淗umans will over time become even more vulnerable to such attacks,鈥� says Saxena.

Reference: