The news
Modulate, a Boston startup whose AI analyzes speech audio directly for tone, intent and voice deepfake detection, said on September 28, 2026, that it raised $25 million led by Future Ventures. The company said the round brings its total funding to $60 million.
Hyperplane and Lakestar also took part, according to the company; SiliconANGLE described both as returning investors. Modulate did not disclose the round's series name or a valuation. Future Ventures co-founder Steve Jurvetson said in the announcement that he sees Modulate ahead technically in audio-native AI, in a market he described as growing fast.
The company's main product is Velma, a platform that listens to recorded or live conversation audio to pick up signals such as emotion, tone, intent, emphasis and synthetic speech. Instead of one large model, it combines more than 100 specialized audio models, an approach the company calls an Ensemble Listening Model. Modulate claims Velma is twice as accurate as traditional large language models at catching true problems and produces seven times fewer false alarms. Chief executive Carter Huffman argued that voice raises problems that "can't be solved from a transcript."
Deepfake detection is central to the pitch. According to SiliconANGLE, Modulate launched Velma Deepfake Detect in March 2026, and the company says its deepfake detection reaches 98.9% accuracy on public benchmarks. TechCrunch reported that Modulate specializes in spotting deepfakes and alerting organizations such as call centers to possible scams, and in checking that AI voice agents follow compliance rules in regulated industries. The company told TechCrunch its technology is also used to monitor cyberattacks carried out through voice calls. SecurityWeek reported that listed uses include protecting healthcare institutions from deepfake attackers.
Modulate says its models process more than 10 million hours of audio a month and have analyzed more than 600 million hours in total. It prices batch transcription at 3 cents per hour of audio, according to SiliconANGLE. Huffman and Mike Pappas founded the company in 2017, SiliconANGLE reported. TechCrunch reported that Modulate began with voice-changing technology for gaming before moving to voice moderation, and that it has 40 to 45 employees and plans to hire about 10 more in the coming months.
The company said it will use the money to grow its team and infrastructure and to expand the models, APIs (application programming interfaces), software development kits, integrations and deployment options it offers to developers and partners building voice applications.
The numbers
- New funding
- $25 million
- Total funding (company)
- $60 million
- Audio processed per month (company)
- More than 10 million hours
- Deepfake detection accuracy on public benchmarks (company)
- 98.9%
- Batch transcription price
- 3 cents per hour of audio
- Employees (TechCrunch)
- 40 to 45
Why CEOs should care
For CISOs and fraud leaders, the phone line is now an attack surface for synthetic voices. Map every workflow where a voice on a call is enough to change something: password resets at the help desk, payment approvals, changes to a supplier's bank details, account recovery in the contact center. Each should have an out-of-band check, such as a callback to a number on file, that does not rely on recognizing a voice. Detection tools can add a layer, but they should not be the only control.
When evaluating vendors such as Modulate, treat benchmark scores as a starting point. A 98.9% accuracy figure on public datasets says little about performance on your own calls, with their accents, line noise and compression. Ask for false positive rates on a sample of your traffic, how fast alerts arrive during a live call, how the tool handles new voice-cloning methods, and what happens to recorded audio, including where it is stored and for how long. Legal teams should confirm that call recording and analysis meet consent rules in every region where customers call from.
For contact center and compliance leaders deploying AI voice agents, tools that analyze tone and intent offer a way to supervise those agents at scale. CFOs should note the pricing signal: when transcription costs 3 cents an hour, the budget question shifts from converting speech to text toward what analysis sits on top of it, and whether that is bought from a specialist or built in-house.
The bigger picture
Huffman's argument is that voice is becoming a main interface for AI, which brings new risks around impersonation and oversight of automated agents. He told TechCrunch that many companies offer transcription but few capture the full nuance of a conversation. Modulate is betting that analyzing the audio itself catches signals a transcript loses, such as tone, emphasis and signs of synthetic speech. Its claims of higher accuracy and fewer false alarms come from the company, and the benchmark rankings it cites are on public datasets, not customer calls.
What’s next
Modulate plans to add about 10 employees to support model building, according to TechCrunch, and to widen its developer tools and partner integrations. The signs to watch are named enterprise customers in banking, healthcare and contact centers, independent testing of its deepfake detection on real call traffic, and whether larger communications and security vendors build similar audio-native analysis into their own platforms.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
