In conjunction with Google AI, Google News Initiative (GNI) is developing a body of synthetic speech ‘Deep Fake’ that contains thousands of phrases spoken by its Deep Learning text-to-speech models. It contains 68 synthetic voices from a broad range of regional accents from English newspaper articles. As malicious actors can synthesize speech in order to bamboozle voice authentication systems, they can even create fake audio recordings to slander public figures.
Deep fakes, audio or video clips created by Deep Learning models that can be exploited for manipulating trust in media. It then becomes complicated to differentiate real from tampered content. For the reason, there was a need for a synthetic speech database. This effort is also in the direction of Google’s AI Principles to make sure strong safety practices to keep away from unintended results that generate risks of impairment. Google Maps and Google Home utilize Google’s speech synthesis or text-to-speech (TTS) technology. Last year, the Google News Initiative announced that it wanted to address deep fakes and other systems that strive to avoid voice authentication systems.
Currently, this dataset is available for participants of the 2019 ASVspoof challenge for developing countermeasures against forged speech. The purpose is to build the Automatic Speaker Verification (ASV) systems to more secure. ASVspoof participants can build systems which learn to differentiate between the real and computer-generated speech by training models on both. The outcomes for this challenge will be announced at the 2019 Interspeech conference in Graz, Austria in September.