Voice Deepfake Detection

Detect synthetic audio.
Protect what's real.

Upload an audio clip and get a real or fake label in seconds. A prototypical meta-learning detector over 3-second chunks powers robust, production-ready detection.

For security teams

Triage suspicious calls and incidents with a fast, explainable deepfake check before escalation.

For journalists & trust groups

Vet leaked audio and submissions with a reproducible detector you can self-host and audit.

For researchers & builders

Experiment with a meta-learning detector and plug its API into your own workflows or dashboards.

Upload

Choose any audio file (WAV, MP3, FLAC). The backend loads it with torchaudio, resamples to 16 kHz, and normalizes for the model.

Predict

A prototypical meta-learning model runs over 3-second chunks and returns P(fake) plus a real/fake label with chunk-level details.

Results

Each run is stored in your browser. View history on Results and clear when needed. Ready for API-backed storage when you need it.

Our research contribution

The golden point: meta-learning layer

What makes our system different is a meta-learning layer on top of a prototypical encoder. Instead of using a fixed classifier, we learn Real/Fake prototypes in an embedding space and train a meta-learner that quickly adapts to new domains or attack types with few examples—improving generalization and robustness to unseen deepfakes. This is our golden point: learning to learn at inference time.

  • Adapts to new domains with minimal data
  • Improves cross-domain and few-shot performance
Read about the meta-learning layer

How it works

From file to label: pipeline and model in detail.

Quick flow

Three steps from upload to result.

1

Upload

Select an audio file. Supported formats include WAV, MP3, and more.

2

Process & infer

We split audio into 3 s chunks, extract features, and run a prototypical meta-learning model.

3

Get label

Receive P(fake) and a real/fake label; results are saved to history.

Our mission

We build and share tools to detect voice deepfakes—so the world can catch harmful synthetic audio with better, advanced, and modern systems. Open source, API access, and a community-driven approach to make the digital world safer and easier to live in.

Read our mission

Developers & open source

API & API keys

We provide API keys and a hosted endpoint (in progress) so you can integrate our service into your apps without hosting locally or on a VPS. Docs show how to use the API; we’ll announce when the endpoint and key signup are live.

How to use the API →

Get started & deployment

Run locally or host on a VPS. Get started for local setup; Deployment covers host on VPS (where to get a server), Nginx, SSL, and security.

Learn more

Deep dive into the pipeline, model, and meta-learning layer.

Ensure the Flask backend is running for Detect and Status. Check Status for health and model info.