Detect AI voice fraud
on live calls.

Anyone can clone a voice in minutes. PulmoForge detects it in real-time — even through Zoom compression, where every other detector goes blind.

Run Demo → Join Waitlist Contact
680%
rise in voice deepfakes
year-over-year in 2024
Pindrop 2025 Voice Intelligence & Security Report
$40B
projected AI fraud losses
globally by 2027
Deloitte Center for Financial Services, 2024
1,300%
increase in deepfake fraud
attacks on enterprises in 2024
Entrust Cybersecurity Institute Report, 2024
1 in 10
adults globally have encountered
an AI voice scam
SQ Magazine, April 2026
Threat landscape
Your meetings are
the new target.

Impostors can slip into everyday virtual conversations — stealing data, misleading teams, and leaving organizations exposed. Continuous participant verification, built to catch synthetic voices as they speak, stops the threat before it spreads.

Enterprise fraud tools cover phone systems and contact centers. PulmoForge is focused on the gap they don't: live video meetings, where codec compression hides what older detectors look for.

Why it matters — the cost of inaction
$25M
Lost in a single deepfake call
A finance worker in Hong Kong was tricked into transferring $25M after a deepfake video call impersonating their CFO — 2024.
11s
To clone any voice
Modern AI needs just 11 seconds of audio to produce a convincing voice clone — available to anyone with a standard device.
$500k
Average incident cost
The average financial loss incurred by businesses from a single deepfake voice fraud attack in past years.
AI voice fraud — growth trend
↑ Rapid growth
Low High 2022 2023 2024 2025 Peak
Where fraud is hitting hardest via Pindrop, 2025
01
Deepfakes in live meetings
Impostors in virtual meetings — stealing data and misleading teams in real-time.
02
Contact center deepfakes
AI-generated voices bypassing agent verification within seconds of connection.
03
Threat monitoring gaps
False positives and missed detections cost enterprises millions annually.
How it works
Simple. Fast. Built for real calls.
01
Someone joins your call
Could be a real person. Could be an AI-generated voice cloned from a CEO, a colleague, or a family member. They sound identical.
02
PulmoForge listens
Audio runs through the detector in real-time, scoring every second of speech — looking for the residual artifacts synthetic voices leave behind, even after compression.
03
You get a risk score
A continuous score tells you how likely the voice is AI-generated. PulmoForge assists human judgment — it doesn't replace it.
Competitive positioning
How we compare to other detectors.
SystemMeeting audioCompression trainedStatus
PulmoForge ✓ Zoom + Opus ✓ Yes ⟳ Rebuilding
Legacy detectors ✗ Mostly lab audio ✗ Rarely Modern gen AI bypasses
Enterprise fraud tools ✓ Phone systems Unknown Closed platforms

Most voice fraud systems were not built around Zoom compression pipelines. PulmoForge was trained directly on compressed meeting audio — this advantage is preserved through the rebuild.

Technical capabilities
What the system does.
Codec resistance
Codec-resistant detection
Trained on Zoom's Opus 32kbps compression pipeline. Detects synthesis artifacts that survive real-world codec degradation — where most detectors fail.
False positive rate
Low false positive rate
Designed to minimize false accusations on genuine voices. Results are probabilistic — intended to assist, not replace, human review.
Architecture
WavLM backbone
Fine-tuned on large-scale speech data. Attention pooling isolates subtle synthesis artifacts across codec-compressed audio. Architecture unchanged through the rebuild.
Inference mode
Live call monitoring
Per-second risk scoring on real-time calls and meetings. Will return with continuous risk explanations post-rebuild.
Early access
Be first when we launch.

The PulmoForge model is being rebuilt on a clean, fully verified dataset. Free beta access goes to the waitlist first — no spam, just a single email when it's ready.

Run Current Demo → Join Waitlist