Blog

Google joins efforts in the effective detection of fake audios

Technological advances are allowing, among other aspects, that voice synthesis technologies can offer increasingly realistic and human voiceovers, being able to arrive at a point where a voice announcement coming from a human staff can be indistinguishable from a computer system.

The bad thing is, taking advantage of that point of realism in the synthesized locutions, there may be malicious people or groups that can misuse these technological advances.

Can you imagine creating an audio clip to discredit a public figure, putting something in its mouth that it didn't really say? Or that they take advantage of these advances to control connected devices through the voice, or even to deceive voice authentication systems?

Well these and other possibilities are there. To prevent them, Google has already advanced during the past year, which, through the Google News Initiative, would offer a set of data that will allow progress in investigations of false audio detection.

And today, the company delivers on that promise. As they point out, the Google News Initiative has partnered with Google IA to be able to make available a set of thousands of phrases pronounced by your voice synthesis models based on deep learning systems, with phrases from English newspaper articles featured through 68 synthetic voices, covering a variety of regional accents.

This dataset will be available to all researchers participating in the ASVspoof 2019 challenge, with the aim of look for mechanisms that enable systems Automatic Speaker Verification (ASV) to be able to effectively distinguish locutions from real people from those artificially generated by computer systems.

Back to top button

Ad blocker detected

You must remove the AD BLOCKER to continue using our website THANK YOU