Voice authentication and security startup Pindrop Security Inc. said today it has raised $100 million in debt financing from Hercules Capital Inc. to accelerate the development of its “cutting-edge” audio, voice and artificial intelligence technologies.

Pindrop is the creator of voice authentication technology that’s used by companies to detect fraudsters and authenticate customers in a secure way at scale. It says the technology has numerous applications, the most popular being to secure contact centers. There, it analyzes customers’ voices and compares them with a comprehensive risk database to verify that callers are who they claim to be.

Other applications include enabling mobile application voice identity-based logins, providing a more convenient way for users to log in to their favorite apps, and consumer devices such as cars, which can be programmed to unlock only for their owner, or access to a playlist on a mobile device.

The company says its technology is widely embedded in contact centers, which are becoming increasingly vulnerable to more sophisticated cyberattacks. It cites studies that show how contact center fraud has increased by more than 60% in the last two years, reaching its highest level since 2019. According to one study, one in every 730 calls to a contact center is expected to be fraudulent, highlighting the need for secure voice authentication tech to eliminate a fraud risk that’s estimated at $5 billion annually.

Pindrop says it serves some of the largest financial institutions in the U.S., including 11 of its largest insurers and eight of the top 10 banks and credit unions. It has analyzed more than 5.3 billion calls for its customers in total, detecting more than 104 million spoof calls, and preventing them from losing an estimated $2 billion to sophisticated fraud scams.

Fighting deepfakes

Having conquered the contact center, Pindrop is now looking to use its technology to aid in the fight against so-called deepfakes, which are AI-generated videos, images and audio files that mislead users by making individuals do or say things that they didn’t actually do. Some deepfakes can be incredibly realistic, and the problem is often described as one of the major challenges in the AI industry.

To combat deepfakes, Pindrop this year debuted a new tool called Pindrop Pulse that’s designed to analyze audio to identify fake videos through voice recognition. The tool can also be used in advanced liveness tests for verification use cases such as Know Your Customer checks.

Pindrop Chief Product Officer Rahul Sood told SiliconANGLE that deepfakes have become alarmingly common because there are more than 330 publicly available text-to-speech engines online that anyone can use to create a fake voice that imitates a real person. He said the challenge is exacerbated by the fact that deepfakes are evolving as more powerful AI models come available, and so it requires a solution that can evolve at the same speed.

“We’re confident that our liveness detection approach is the right one,” Sood said. “Humans are notoriously inefficient at catching deepfakes, with studies showing that our accuracy is between 38% and 54%, depending on the person. Voice recognition systems are slightly better, with accuracy of 38% to 88%, but the potential for false positives means they are not quite there yet.

According to Sood, Pindrop’s deepfake detection tool Pindrop Pulse is based on the belief that “bad AI” can be beaten by “good AI.” It’s based on a liveness detection engine that’s powered by deep neural networks, and has demonstrated accuracy of up to 94.6% in independent testing, he said, with a false positive rate of just 1%. “This accuracy goes up to 99% when it’s combined with voice recognition, device recognition and behavioral analysis,” he explained.

The CPO added that Pindrop has been working on identifying deepfakes for more than eight years. The Pindrop Pulse offering is trained on a vast and evolving dataset that’s made up of more than 22 million utterances and able to identify more than 330 text-to-speech engines, he said. What’s more, the company is constantly evolving and updating Pindrop Pulse to stay ahead of the most sophisticated fraudsters.

As an example of the company’s prowess in identifying deepfakes, Pindrop says it identified the specific text-to-speech engine that was used in the President Biden robocall attack earlier this year. In that attack, a number of individuals received robocalls purportedly from Biden himself, urging them not to vote in New Hampshire’s Democratic primaries.

The company is so confident in the ability of Pindrop Pulse to detect deepfakes that it followed that launch with its Pulse Deepfake Warranty offering, which compensates eligible customers if its tools are unable to detect a deepfake or synthetic voice fraud, resulting in financial losses to the affected customer.

Sood clarified that this is not an insurance policy because it does not cover all known situations that might lead to fraud or financial losses. Rather, it’s focused exclusively on the performance of Pindrop Pulse, and enables customers to claim a reimbursement in the rare instance that the tool fails to catch a synthetic voice.

“In many ways, the warranty is more like a bug bounty, where we are asking our customers to tell us about the rare instances when our solution does not work,” Sood said. “Then we can use that information to update our models so that our solution grows stronger over time.”

One of Pindrop’s most enthusiastic backers is John Chambers, former chief executive officer of Cisco Systems who now serves as the CEO of JC2 Ventures, which invested in the company in a prior funding round.

Chambers predicted last year that the issue of deepfakes would become the most critical AI concern for enterprises in 2024, and he says that forecast has proven to be very accurate.

“In my career, I have had the opportunity to help position companies to lead in these big market transitions, and I can say with confidence that Pindrop is uniquely positioned to do just that with its revolutionary voice security and authentication technology,” Chambers said.

He added that he believes Pindrop will “achieve the fastest run rate to $5 million in an entirely new product area by an existing company, compared to any other company I have seen.”

Image: SiliconANGLE/Microsoft Designer