Clement Brako Akomea Table of Contents
Modulate deepfake detection has received a major funding boost, as the AI firm raised an additional $25 million from Future Ventures, with participation from Hyperplane and Lakestar.
The round brings Modulate’s total funding to $60 million, SecurityWeek reported on September 28, 2026. The company builds AI models that analyze human voice in real time.
The money targets a growing problem: fraud and abuse carried out through synthetic and manipulated speech, which is hard to catch without AI.
Modulate Deepfake Detection: Key Takeaway
- Modulate raised $25 million, lifting total funding to $60 million, to expand its real time voice AI models for spotting deepfakes and abusive conversations.
Voice fraud often ends with a stolen identity or a compromised account. These tools help reduce that risk:
- GetTrusted: identity and trust verification for organizations that need to confirm who they are dealing with.
- Optery: personal data removal, which limits the details a scammer can use to impersonate you.
- 1Password: a password manager that keeps every account on a strong, unique password.
Who Is Modulate and What Does It Build?
A company focused on listening, not speaking
Modulate describes itself as a frontier AI firm. It builds AI models designed to help machines understand the nuances of human conversation, whether the speech is natural or fake. According to the SecurityWeek report, the company uses AI to analyze voice rather than to create it.
That distinction matters in a market crowded with tools that generate synthetic voices. Modulate sits on the defensive side of the technology, where the goal is to recognize what is being said, how it is being said, and whether the voice is real.
The Ensemble Listening Model architecture
Modulate uses a proprietary design called the Ensemble Listening Model, or ELM. ELM generates the individual AI models that power the company’s Velma platform. Velma brings together more than 100 of these specialized models.
Together, those models detect and interpret several kinds of signals in audio: emotion, tone, intent, synthetic speech, and conversational behavior. Synthetic speech is voice produced or altered by AI, and it is the raw material of a voice deepfake.
Why several signals matter
A cloned voice may sound convincing, yet the behavior around it can still look wrong. Because Velma reads emotion, intent and conversational behavior alongside synthetic speech, it can weigh more than one clue at once. The source article does not describe how the models are combined, so readers should treat this as a general description of the approach.
Scale and performance claims
Modulate says it currently analyzes more than 10 million hours of audio each month, and more than 600 million hours in total. The company also states that its transcription and deepfake detection models both ranked first on public benchmarks such as Hugging Face.
Modulate further claims that Velma delivers results twice as accurate as traditional LLMs (large language models, the text based AI systems behind many chatbots) while producing seven times fewer false positives. A false positive is a genuine call or speaker wrongly flagged as fake or abusive. These figures come from Modulate. The SecurityWeek article does not cite independent testing of them.
Funding Details and Planned Use
Investors and totals
The new $25 million came from Future Ventures, with participation from Hyperplane and Lakestar. It lifts Modulate’s total funding to $60 million.
How Modulate plans to spend the money
Modulate says the funds will develop its team and infrastructure. It will also expand the models, APIs, SDKs, integrations and deployment options available to the developers and partners who build voice applications. An API is a defined way for one piece of software to request a service from another, and an SDK is a toolkit that helps developers build with that service.
The company will also keep investing in its own underlying research and technology. In practice, the plan is to make audio analysis something developers can plug into products instead of building from scratch.
Why Voice Deepfakes Are a Growing Security Problem
Voice is becoming a main interface for AI
Modulate’s CEO, Carter Huffman, argues that voice is turning into a primary interface for AI. He says that shift creates problems that a transcript alone cannot solve. A written transcript loses tone, emotion and the acoustic traces that can reveal a synthetic voice.
For security teams, that gap is significant. A fraudster who imitates a trusted voice may say words that look harmless on paper. Only the audio itself carries the evidence that the speaker is not who they claim to be.
AI assisted fraud is now routine
The SecurityWeek article states that adversarial use of AI assisted fraudulent deepfakes, along with simple misuse of language for abuse, is now a fact of life. A deepfake is media, such as audio or video, that AI has generated or altered to look or sound like a real person.
The article adds that detecting deepfakes and recognizing abuse in time to prevent harm is almost impossible without AI. That is the case Modulate is making: human listeners cannot keep pace with the volume and quality of manipulated voice.
For a related look at how scammers already use phone calls, see our guide to vishing attacks and how to prevent them.
Where Modulate Deepfake Detection Is Used
Modulate lists several typical applications for its technology. They fall into three groups.
Security and fraud protection
Protecting healthcare institutions
One use is defending healthcare institutions from deepfake hackers. Hospitals and health services handle sensitive records and often rely on phone contact, which makes a convincing fake voice a real risk.
Shielding agents from voice unmasking
Modulate also says its technology can protect agents in risky scenarios from being identified through advanced voice masking. The source does not expand on this use, so no further detail is given here.
Trust and safety on online platforms
Reducing extremism and harassment
Modulate applies its models to reduce extremism and harassment on social platforms. Voice chat is hard to moderate because abuse happens live and leaves no text to scan.
Stopping child grooming conversations
Another listed use is to detect and stop voice conversations involving child grooming. Because Velma works in real time, the aim is to intervene during the conversation rather than review it afterward.
Voice AI agent quality
Improving emotion and empathy
Developers can use the platform to enhance the emotion and empathy capabilities of voice AI agents. Here, an AI agent means software that holds spoken conversations with people.
Monitoring performance
Modulate also supports observing and monitoring how voice agents perform. Its CEO said organizations need to check whether their voice agents are actually working the way they are supposed to.
Since Velma operates in real time, applications built on it can act while a conversation is still happening. SecurityWeek has covered the wider market too, including isVerified emerging from stealth with voice deepfake detection apps.
What Modulate’s CEO Said
Huffman said Modulate is already using audio native AI to protect organizations from deepfake attacks, help voice agents understand emotion and respond with more empathy, identify dangerous behavior in online conversations, and monitor voice agent performance.
He also said developers should not have to rebuild the audio intelligence layer each time they create a new voice experience. Modulate’s stated mission is to supply the models and infrastructure so developers can concentrate on the application they want to build.
Huffman described the opportunity for audio native AI as expanding quickly. He said the company has proven its technology at scale and that the investment will let it grow the team and move faster to meet demand.
Implications of Voice AI Abuse and Detection
The funding is a business story, but it points to shifts that security professionals should consider. The points below are analysis based on the facts reported by SecurityWeek.
Implications for healthcare and regulated organizations
Healthcare institutions appear in Modulate’s list of protected settings for a reason. Staff in clinics and administrative offices regularly act on spoken requests, such as confirming a patient, releasing information or changing an account detail.
If a caller can imitate an authorized person, a single successful call may expose records or trigger a harmful change. Detection that runs during the call gives an organization a chance to pause the request, which a later review of a recording cannot do.
Implications for developers building voice products
Modulate’s plan to expand APIs, SDKs and integrations shows where the market is heading. Detection is being packaged as a component that developers add to their own voice applications.
That lowers the barrier to building safer products. It also raises expectations: as detection becomes easier to add, users and regulators may increasingly expect voice products to include it.
Implications for trust and safety teams
Moderating voice is harder than moderating text. There is no message to search, and harmful behavior can pass in seconds. Tools that analyze emotion, tone, intent and conversational behavior give moderators signals that a transcript cannot provide.
The uses Modulate lists, from harassment to child grooming, show that the same technology serves both security and safety goals. For platforms, that means a single investment can support both fraud controls and user protection.
Implications for security teams and everyday users
Deepfake voice fraud rarely works alone. It often pairs with phishing messages, spoofed email and stolen credentials. Defenders need layered controls: strong password hygiene, email authentication and clear procedures for verifying unusual requests. Our overview of how to stay safe from phishing scams covers the basics.
Individuals can also shrink their exposure by limiting how much personal information is available online, since scammers use it to make impersonation believable. The trend in AI powered fraud prevention platforms shows that defenders are adopting the same class of tools.
Limits and open questions
Several questions remain. The performance figures come from Modulate itself, and the source article does not mention independent verification. Detection is also a moving target, since attackers adapt as defenses improve.
The SecurityWeek article describes an arms race in broad terms, and its related coverage, including a piece on the AI arms race between deepfake generation and detection, explores that contest further. Buyers should test any detection tool against their own call conditions before relying on it.
This section contains affiliate links; we may earn a commission at no cost to you.
Deepfake fraud rarely arrives alone. Strengthen the other layers of your defense:
- EasyDMARC: email security and DMARC, which helps block spoofed messages that often accompany impersonation scams.
- CyberUpgrade: cybersecurity compliance and management for organizations formalizing their controls.
- Passpack: a password manager for keeping team and personal credentials organized and unique.
Looking Forward
Modulate’s $25 million round lifts its total funding to $60 million and gives it more room to expand its Velma platform, its developer tools and its research.
The company positions real time voice analysis as a way to stop harm while it is happening, across fraud, harassment, child safety and voice agent quality.
The performance claims still need independent testing, and attackers will keep adapting. Even so, the round shows investors betting that audio analysis will become a standard layer of protection for voice applications.
Questions Worth Answering
How much money did Modulate raise?
- Modulate raised an additional $25 million, bringing its total funding to $60 million.
Who invested in the new round?
- The funding came from Future Ventures, with participation from Hyperplane and Lakestar.
What does Modulate do?
- It builds AI models that analyze voice to understand emotion, tone, intent and whether speech is synthetic, rather than generating voice.
What is the Velma platform?
- Velma is Modulate’s platform, which combines more than 100 specialized models created by its Ensemble Listening Model architecture.
How much audio does Modulate analyze?
- The company says it analyzes more than 10 million hours of audio each month and more than 600 million hours in total.
What accuracy does Modulate claim?
- Modulate says Velma is twice as accurate as traditional LLMs and produces seven times fewer false positives. These are company claims.
Why does real time analysis matter?
- Because Velma operates in real time, applications built on it can intervene during a conversation instead of reviewing it afterward.
What will the new funding be used for?
- Modulate will grow its team and infrastructure, expand its models, APIs, SDKs, integrations and deployment options, and continue its research.
What are the main use cases?
- They include protecting healthcare institutions from deepfake hackers, improving voice agent empathy, reducing extremism and harassment, monitoring voice agents, stopping child grooming, and protecting agents from voice based identification.
Affiliate links: protect more of your digital life with Tresorit for encrypted storage, Tenable for vulnerability management, and IDrive for backup and ransomware recovery.
Sources:
