Modulate raises $25M Series B for voice AI models
What's the deal? Boston-based Modulate has raised US$16.2M Series B to develop its AI-powered voice analysis platform. Future VenturesDealroom has a profile for this one. Try Dealroom → led the round, with Hyperplane and Lakestar participating.
What does it do? Modulate gives businesses tools to transcribe calls, read emotions, detect voice deepfakes and AI-generated music, and monitor whether voice agents comply with rules in regulated industries. It flags fraud for call centres, assesses AI agents' responses to customers, and tracks cyberattacks conducted over voice calls.
How it works: The platform runs more than 100 models split into two groups. One extracts signals from audio — emotion, tone, language, and whether a voice is synthetic; the other reads intentions, such as what a customer wants or whether a caller is trying to deceive.
Why small models? Chief technology officer Carter Huffman says smaller models avoid the need for specialised hardware or heavy compute. The company can train new capabilities, add them, and call on them through an orchestrator.
Background: Modulate was founded in 2017 by Mike Pappas and Huffman, who met studying physics at the Massachusetts Institute of Technology. It started with voice-changing tools for gaming before pivoting to voice moderation, and now works on detecting AI-generated audio and analysing speaker intent. According to PitchBook, it had raised $41 million before this round at US$110M valuation.
What's the money for? Modulate, which employs between 40 and 45 people, plans to hire another 10 specialists for model development. It is also expanding on-premises and on-device deployment to improve privacy.
The signal: The round sits at roughly the 29th percentile by size — modest against the sector's headline raises. But the pitch reflects a wider shift: as AI-generated voices proliferate, tools that can detect deepfakes and audit voice agents are becoming a distinct market of their own.
Read more: ua.news
Image credit: comedy_nose
