Modulate Cuts Synthetic Voice Detection Time to 1.6 Seconds

Daniel Okafor

Modulate says its latest models detect AI-generated speech in a median of 1.6 seconds. The figure is down from 2.25 seconds. Accuracy remains above 99.1% on real-world AI and human call evaluations. Modulate announced new investments from Twilio Ventures and Kayak co-founder Paul English. The company did not disclose the size of those additional investments.

The backing follows a $25 million financing round announced in September. Future Ventures led the round, with Hyperplane and Lakestar participating. Modulate is also adding more than a dozen audio-native signals to its Velma platform. The signals aim to help enterprises identify who, or what, is calling and understand how an AI-led conversation unfolds.

What does the new backing fund?

Modulate said the new capital will support research, product development, developer adoption, partnerships and broader deployment options. September’s announcement put the same priorities at the center of its expansion beyond earlier voice-safety applications. Both releases bring detection and supervision closer together: enterprises can recognize generated speech, examine surrounding behavior and assess whether their own voice agents meet intended goals.

What can businesses do in 1.6 seconds?

Modulate says the faster median gives businesses time to identify synthetic speech, route a call, flag it for review or apply additional verification while the conversation continues. Velma examines underlying audio without relying on watermarks inserted by individual voice-generation providers, including for newer speech-to-speech and full-duplex models.

The improvement from 2.25 to 1.6 seconds represents about a 29% reduction in Modulate’s reported median. It does not guarantee that every call will receive a verdict within that interval or that a complete security workflow will finish so quickly. The announcement also omits the evaluation sample size and any breakdown of false positives and false negatives behind the greater-than-99.1% accuracy figure.

Carter Huffman, CEO and co-founder of Modulate, said: “Call operators – human or AI agent – must be able to recognize what’s happening in those conversations and respond appropriately while the interaction is still ongoing.”

For contact centers, Modulate says an early warning can prompt intervention before the conversation ends. A warning after an account change or sensitive disclosure serves a different purpose from a signal available while the interaction continues.

How does Velma analyze a live call?

Velma uses what Modulate calls an Ensemble Listening Model, or ELM: an architecture that coordinates specialized models rather than relying entirely on a single general-purpose model to interpret a transcript. More than 100 models cover signals including:

• Emotion and tone
• Intent and cadence
• Synthetic speech and conversational behavior

An orchestration layer combines lower-level signals into higher-level assessments while preserving links to the underlying evidence. Time-stamped events can indicate where a conversation changed, giving operations teams more specific material than an unexplained call-level risk score.

A transcript captures words, but it does not preserve every pause, interruption, change in emphasis or delivery pattern. Modulate says Velma adds acoustic information so developers can examine how a conversation proceeds alongside what participants say.

The company’s deepfake-detection documentation describes acoustic classification of audio segments with time boundaries, verdicts and confidence scores. The API supports recorded files and live streams. Modulate says a confidence score supports a classification but does not prove a caller’s identity or intentions.

Does synthetic speech prove fraud?

Modulate’s voice-fraud screening guide separates deepfake detection from conversation analysis. Deepfake detection evaluates whether a voice is synthetic, while conversation analysis examines behavior. A legitimate AI assistant may use generated speech, while a human caller may attempt social engineering without using voice synthesis.

Modulate says that distinction matters for people using assistive speech technology. A synthetic-voice flag alone does not provide sufficient reason to treat a caller as malicious. The new release allows Velma to combine that flag with intent, manipulation, tone and conversational-behavior signals to support more informed handling.

The company describes a graduated response:

• Ordinary routing for permitted automation
• Additional verification where a transaction warrants it
• Escalation when the conversation shows suspicious behavior

Modulate says the detection system supplies evidence, while the customer’s policies determine the response.

How can companies supervise their own AI agents?

The new Velma signals also address a commercial problem that fluent delivery can conceal: an AI agent may sound polished while providing a poor service experience. Modulate says the additions help evaluate response timing, audio quality and conversational misunderstandings that transcript-only analytics can miss.

On its AI guardrails page, Modulate describes Velma as monitoring participants rather than acting as the conversational agent itself. Customers can configure monitoring around operating procedures and expected behavior, then send signals into existing systems through APIs and webhooks.

This separation lets teams investigate where an agent causes frustration, goes outside policy or fails to respond appropriately. Teams can then use those findings to improve deployment.

What does Twilio’s backing change?

Twilio Ventures describes its focus as AI, developer tools and customer-engagement businesses, making audio intelligence relevant to its investment remit. Modulate already had a technical connection with Twilio. A May 2025 announcement described compatibility for VoiceVault and ToxMod, including Twilio Media Streams-compatible APIs that bring live call audio into Modulate’s systems.

The new investment builds on that history. Modulate said the announcement did not establish a new exclusive distribution agreement with Twilio. For enterprise buyers, the next test is whether Velma’s faster signals can drive intervention during a live call without treating legitimate generated speech as fraud.