Loading page…
The funding backs new developer tools and industry-specific models. Modulate’s pitch is that the sound of a call can reveal things its transcript misses.
Listen to this story
Modulate’s $25 million round, led by Future Ventures, will fund SDKs, APIs and industry-specific models to extend Velma’s audio analysis into more developers’ products; those offerings are planned, not launched. The platform combines more than 100 specialized models for signals including synthetic voice, caller intent and possible dissatisfaction, and the company reports processing over 10 million audio hours monthly. Modulate claims up to 1,000× efficiency versus one large model, but that is not an accuracy result
An existing batch-transcription API costs $0.03 per hour; the broader SDK and API package is still planned.
Modulate has about 40–45 employees and aims to hire roughly 10 more in the coming months to build models.
PitchBook cited $41 million in pre-round funding and a prior $170 million valuation; neither figure is the new round’s valuation.
A transcript can show what a caller said. Modulate wants its software to catch what the words leave out, from signs of a synthetic voice to a customer’s dissatisfaction. The company has raised $25 million to widen access to that voice-analysis technology, with Future Ventures leading the round and Hyperplane and Lakestar participating.
Modulate’s Velma platform works from conversation audio, rather than relying only on the words produced by transcription. Its models examine features such as tone and vocal emotion, alongside signs that a voice may be synthetic. Other models assess intent: what a caller is trying to convey or whether a conversation suggests a scam or rule violation.
Instead of assigning all those jobs to one large model, Modulate uses more than 100 smaller, specialized models. Its system selects models for a task and combines their results. Co-founder Carter Huffman told TechCrunch that this approach lets the company add capabilities without depending on specialized hardware or heavy computing resources. Modulate says the design can be up to 1,000 times more efficient than sending the same audio to one large model; that comparison is a company claim, not a measure of how accurately it understands any particular call.
Modulate sells more than deepfake detection. Its tools can alert a call center to a possible scam and check how an AI voice agent responds to customers, including whether it follows applicable rules. The analysis can sit beside a company’s existing voice system rather than replace it. That gives the company a role in judging calls handled by both people and AI agents.
Huffman offered a less obvious customer-service problem: someone may stay polite with an AI agent while leaving the call dissatisfied. A simple positive-or-negative emotion score could miss that difference. Modulate’s aim is to give businesses a more detailed account of what happened, though recognizing dissatisfaction from a voice remains an interpretation of the call, not a direct statement from the customer.
Modulate says its models process more than 10 million hours of audio a month.
Modulate puts its lifetime processed-audio total above 600 million hours.
The funding is intended to put those models in front of more developers. Modulate plans new software development kits and APIs—tools that would let other teams build its capabilities into their own software—as well as models tailored to particular industries. Those tools are plans, not part of a newly released developer package. An existing API already offers batch transcription at three cents an hour.
Modulate is also working to make more of its technology available on a customer’s own premises or device. That matters for buyers who want voice analysis without sending every call through a remotely hosted service, but the company has described the wider deployment options as work in progress. It currently has about 40 to 45 employees and aims to add roughly 10 more in the coming months to build models.
Founded in 2017 by Huffman and Mike Pappas, Modulate first worked on voice modulation for gaming, then moved toward moderating voice chat. Its current pitch reaches call centers, security teams and businesses running voice agents. PitchBook data cited by TechCrunch put its funding before this round at $41 million and its earlier valuation at $170 million; neither figure is a valuation for the new financing.
The more consequential test is whether those judgments are useful when they are wrong as well as when they are right. A flagged deepfake or a supposedly dissatisfied caller can prompt a business to act. The funding gives Modulate room to develop the system, but the reported audio volume and efficiency claim do not establish how reliably it makes those distinctions in customer calls.
Loading discussion...
Join the conversation
When would that help, and when might it misread someone?
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.