A Boston-based startup specializing in voice intelligence has closed a $25 million funding round to expand its platform, which relies on a suite of small, purpose-built AI models to deliver capabilities including transcription, emotional analysis, deepfake detection, AI-generated music identification, and policy enforcement for voice agents operating in regulated industries.

The company, Modulate, is part of a broader wave of voice AI businesses attracting investor interest - particularly those focused on making automated voice systems more human-sounding and contextually aware. The startup also competes in adjacent spaces, including companies working to decode the intent behind human speech and those building defenses against the increasingly accessible practice of voice cloning and deepfake audio.

The round was led by Future Ventures, with additional participation from Hyperplane and Lakestar. According to data from PitchBook, Modulate had previously raised approximately $41 million in total funding at a valuation of around $170 million before this latest injection of capital.

From Gaming Roots to Enterprise Voice Intelligence

Modulate was founded in 2017 by Mike Pappas and Carter Huffman, two former MIT physics undergraduates who crossed paths during their studies. The company's earliest work centered on voice modulation tools for the gaming sector, helping players alter their voices in real time during online play. Over time, however, the company shifted its attention toward voice-based content moderation, developing tools designed to identify harmful or policy-violating speech within gaming environments and other digital platforms.

With the rapid proliferation of voice AI systems across industries, Modulate has once again evolved its focus. The company now concentrates heavily on detecting various forms of AI-generated audio and on analyzing the underlying intent within human conversations - capabilities that carry significant commercial and regulatory weight as AI-driven customer interactions become more prevalent.

"Our insight into the voice AI space is that a lot of folks are doing transcription, but there's not really any capability out there that gets the full nuance and full understanding of a conversation, which is so important when you're talking to another human being," said co-founder Carter Huffman.

A Two-Tier Model Architecture Built for Scale

Modulate currently operates more than 100 individual AI models, which the company organizes into two broad functional categories. The first group consists of signal extraction models, designed to interpret vocal characteristics such as emotion, tone, language, and whether a voice is human or synthetically generated. The second group comprises analysis and detection models, which focus on understanding intent - including what a caller is trying to accomplish, whether they are violating applicable policies, or whether they may be attempting to defraud the person or organization on the other end of the line.

Huffman noted that the decision to build around smaller, specialized models rather than large monolithic systems provides meaningful operational advantages. Because the models are compact, Modulate does not require specialized hardware or significant compute resources - a consideration that becomes increasingly relevant as the cost of AI inference continues to rise. The architecture also makes it easier to train new models with additional capabilities, slot them into the existing system, and have an orchestration layer call upon them as needed for any given task.

Key Use Cases: Fraud Detection, Compliance, and AI Agent Monitoring

Modulate serves a diverse range of enterprise customers, but several core use cases define its current commercial focus. The company is particularly active in deepfake detection, alerting organizations such as call centers when an incoming call may involve a cloned or synthetically generated voice - a capability that has grown in importance as voice cloning tools have become widely accessible to bad actors.

Beyond fraud detection, Modulate's platform is also used to evaluate the performance of AI-powered customer service agents. As enterprises increasingly deploy automated voice systems to handle customer interactions, the need to assess call quality and ensure regulatory compliance has grown substantially. Modulate's tools can monitor how AI agents respond during calls, flagging potential compliance violations and assessing whether the interaction met applicable standards. Because of this positioning, the company's platform typically operates alongside an enterprise's existing voice stack, functioning as an analytical layer rather than a replacement for the underlying communication infrastructure.

The startup has also disclosed that its technology is being applied in the context of cybersecurity, specifically to detect and monitor cyberattacks conducted through voice channels - a threat vector that has grown alongside the broader adoption of voice-based communication in enterprise settings.

Rethinking Emotional Analysis in Customer Interactions

As AI-driven customer service becomes more common, enterprises face growing pressure to understand not just whether a call was completed, but why it succeeded or failed. Modulate argues that conventional approaches to sentiment analysis fall short of providing that level of insight. Basic positive-negative-neutral classifications, the company contends, miss the subtleties that often determine whether a customer will remain loyal or churn.

Huffman illustrated this with a specific example: customers interacting with AI agents or automated bots frequently maintain a polite or neutral tone even when they are deeply dissatisfied with the outcome of a call. Standard sentiment tools, which rely heavily on vocal affect and word choice, may classify such interactions as successful when they are anything but. Modulate's platform is designed to go deeper, providing enterprises with more granular data about customer intent and satisfaction that goes beyond surface-level emotional cues.

"I think when companies think of emotion analysis, they think if the customer was neutral or positive, the call was a success, and if the customer was negative, the call was a failure. But actually, many times people will be polite even to AI agents or bots. And they won't come across as angry, but they'll be very dissatisfied," Huffman explained.

Growth Plans and What Comes Next

Modulate currently employs between 40 and 45 people and plans to hire approximately 10 additional staff in the near term, with the new roles focused primarily on model development and engineering. The company is also working to expand its on-premises and on-device deployment options, a move intended to address growing enterprise demand for solutions that keep sensitive voice data within controlled environments rather than routing it through cloud infrastructure.

The push toward local deployment reflects a broader concern among regulated industries - including finance, healthcare, and legal services - about data privacy and the risks associated with transmitting sensitive customer conversations over external networks. By strengthening its capabilities in this area, Modulate aims to deepen its foothold in sectors where compliance requirements make data residency and processing locality a critical factor in vendor selection.

With the new capital in hand and a platform that spans detection, analysis, and compliance monitoring across voice interactions, Modulate is positioning itself at the intersection of several fast-moving trends: the proliferation of AI-generated audio, the expansion of automated customer service, and the growing regulatory scrutiny of how enterprises handle voice-based communications.