Pindrop has won the Voice Cloning Challenge run by the U.S. Federal Trade Commission (FTC), along with three smaller organizations.
The challenge results were announced just as investors in India were viewing recommendations on stock purchases that appeared to be from National Stock Exchange (NSE) CEO and MD Ashishkumar Chauhan. The fraudulent video, shared on social media, faked both the face and voice of Chauhan and the NSE logo. The NSE has issued a public caution about the deepfake reminding investors of the sources for legitimate information.
The submissions to the FTC challenge from smaller organizations, AI Detect, DeFake and OriginStory, will split $35,000 in prize money. AI Detect was submitted by OmniSpeech, OriginStory by a startup of the same name, and DeFake was submitted by Washington University (St. Louis) Department of Computer Science and Engineering Assistant Professor Dr. Ning Zhang.
Pindrop and AI Detect both utilize AI algorithms to detect characteristics that reveal a given voice sample is synthetic. Pindrop’s submission makes use of the biometric liveness detection technology the company provides to commercial customers, while AI Detect’s also includes a framework for the responsibilities of public and private sector organizations. Pindrop says its technology works in real-time, with speech samples recorded over a phone (8 kHz) or digital channel (16 kHz).
Pindrop’s technology was also judged the only effective one among several deepfake detection tools tested recently by NPR.
DeFake proposed a technique of watermarking voice samples that could be used in the generation of voice clones to prevent accurate cloning, without being detectable by humans in the source sample. OriginStory collects signals from sensors commonly contained in electronic devices to measure speech acoustics and “co-occuring biosignals in the throat and mouth” to authenticate that genuineness of the sample and create a watermark that attests that it is real.
The challenge was originally unveiled last year with $25,000 in prize money available to smaller organizations.
The FTC has proposed new rules on impersonation fraud and stated the jurisdiction of the Telemarketing Sales Rule over AI fraud calls, as additional measures to combat voice deepfakes.
“When it comes to AI-driven fraud, the FTC will continue using every tool to deter harmful practices, shut down bad actors, and spur innovative proposals to help protect consumers,” says Samuel Levine, director of the FTC’s Bureau of Consumer Protection.
Veridas launches voice fraud detection software
A new technology has been launched by Veridas to combat voice fraud, whether in the form of pre-recorded audio, manipulated or artificially generated speech, or deepfakes.
Voice Shield software can analyze millions of calls in just milliseconds, according to the announcement, regardless of language or text spoken, and make judgements from 3 seconds of audio data, without a reference sample. Without a need for prior registration, the software also helps optimize conversion rates, the company says.
Veridas’ technology was developed based on its experience in securing 5 million processes globally with voice biometrics, and the company cites a ten-fold increase in deepfakes detected from 2022 to 2023.
“Our journey with this technology is not new. We’ve been crafting and refining these solutions for years as a vital part of our core offerings,” states Eduardo Azanza, CEO of Veridas. “Now, in the face of the fraud avalanche coming from deepfakes, we believe it is the perfect timing for us to launch Voice Shield as a standalone product as well to protect both companies and end-users from fraud.”
Article: Pindrop, veridas raise biometric defenses against voice fraud