Pindrop, a company specializing in detecting deepfakes, secures a $100M loan to expand its services.

As artificial intelligence technologies facilitating the creation of deepfakes become more accessible, the hazards associated with them escalate. A striking 245% jump in deepfake incidents globally from 2023 to 2024, partially fueled by upcoming electoral events, was reported by verification service Sumsub. The business sector is not spared, with a Business.com survey uncovering that 10% of businesses have suffered from deepfake-related frauds, including fake voice replication.

This burgeoning trend has fortunately proved beneficial for businesses developing defenses against deepfakes and the technology to generate them. Pindrop, one of these firms, announced a significant financial milestone on Wednesday, securing a five-year, $100 million loan from Hercules Capital. CEO Vijay Balasubramaniyan stated that the funding is designated for product enhancement and team expansion.

“The escalation of generative AI capabilities has notably enhanced the potency of voice cloning,” disclosed Balasubramaniyan in a dialogue with TechCrunch. “As a result, AI-powered deepfake detection tools have become crucial for every call center to remain vigilant against fraudulent activities.”

Specializing in solutions that both thwart deepfakes and support multi-factor authentication, Pindrop targets the banking and finance sector among others. The company prides itself on its capability to verify the identity of contact center callers with a superior level of accuracy than alternative offerings.

“Employing a dataset consisting of over 20 million expressions, both artificial and real, our AI models are trained to discern between authentic human voices and those that are artificially generated,” explained Balasubramaniyan. “Moreover, we’ve developed over 330 text-to-speech (TTS) models to pinpoint TTS systems utilized in crafting deepfakes.”

Detecting deepfakes accurately is hindered by inherent biases in current models. A number of audio models exhibit a bias towards recognizing voices with Western, American accents and falter with varied accents and dialects, potentially mistaking legitimate voices for deepfakes.

The effectiveness of synthetic training data in reducing or amplifying biases in AI models is subject to debate. Balasubramaniyan argues that such data, in fact, helps in reducing bias, with Pindrop’s voice authentication technologies focusing on identifying the unique “acoustic and spectro-temporal features” of a voice rather than its pronunciation or language.

“Discrimination based on tone, accent, and dialect variations by AI-driven voice recognition systems can inadvertently introduce racial biases,” Balasubramaniyan pointed out. “These biases stem from the homogeneity of training datasets, which might not adequately represent diverse ethnic, racial, gender, and other distinctions, hence restricting the spectrum of data on which the AI systems are trained.”

Despite the challenges, Pindrop has achieved considerable success since its 2011 inception by Balasubramaniyan, alongside Paul Judge, a former Barracuda Networks chief research officer, and Mustaque Ahamad. Based in Atlanta, the company, now employing roughly 250 individuals, has garnered $234.77 million in venture funding from entities such as Vitruvian Partners, CapitalG, IVP, and Andreessen Horowitz.

When queried about the preference for debt over equity financing this round, Balasubramaniyan explained that it was a “desirable option” to “secure capital for growth” without diluting current stakes in Pindrop. (A strategy not uncommon.)

Balasubramaniyan added that the loan would facilitate Pindrop’s expansion into new sectors including healthcare, retail, media, and travel.

“The rise of generative AI has surged global demand for our solutions, and we are exploring expansion opportunities into countries particularly vulnerable to deepfake threats,” he mentioned. “Pindrop aims to empower organizations to defend against the increasing menace of fraud and deepfake offenses with our advanced fraud prevention, authentication, and liveness verification technologies.”