Audio deepfake detection that runs on CPU.

Aletheia’s models are small enough to run on standard CPUs, with no GPU required. Detect synthetic and cloned speech on your own servers, in a private VPC or on the device.

AletheiaTruthDeepfake detection

Measured on CPU and GPU.

Current-model latency is supplied by Detectif.ai. The benchmark and handset figures come from the independent Podonos benchmark and our deployment study.

84 ms
On CPU, current model
24 ms
On GPU, current model
97.3%
Deepfakes caught on the independent Podonos benchmark
99.65%
Handset clips processed faster than real time, CPU only, deployment study

Why run deepfake detection on CPU.

No GPU fleet

Run detection on the CPU servers you already operate. Small models need little compute per call, which keeps the cost of running them low.

Next to your calls

Deploy beside the SBC, recorder or telephony stack your calls already pass through, so the score arrives while the call is live.

Audio stays inside

On-premises and on-device deployments keep inference on your network or the handset. No audio has to leave for a hosted GPU service.

Deployment options.

Use the hosted API, deploy in a private VPC, run on your own CPU servers or embed detection on a device.

Where the audio goesAudio crosses into our cloud
Your systems
Your telephonySIP · contact center · application
Encrypted audio
Detectif.ai cloud
Detectif.ai runtimeManaged and scaled by us
Signal response
Your systems
Your policy engineContinue · step up · review
Your infrastructureDetectif.ai model runtime
Hosted API

Start with a managed runtime.

Stream the caller’s audio to Detectif.ai over an encrypted connection. Receive a signal in your application and apply your own decision policy.

Audio processed in
Detectif.ai-managed cloud
Operated by
Detectif.ai
Network path
Encrypted connection to Detectif.ai
Release control
Runtime and model updates managed by us

A straightforward starting point for evaluation and cloud-based call flows.

Study reference: 14.4 ms median per clip on an NVIDIA L4 GPU, FP16.Read the deployment study

CPU deployment questions.

What teams ask before running deepfake detection without a GPU.

Can audio deepfake detection run on a CPU?

Yes. Aletheia, Detectif.ai’s audio deepfake detection model, runs on standard CPUs with no GPU required. The current model scores audio in 84 ms on CPU and 24 ms on GPU.

Does Aletheia need a GPU?

No. A GPU is optional. The same small models run on CPU servers, in a private VPC, on-premises or on-device, and a GPU only lowers latency further.

Can deepfake detection run on-premises on CPU servers?

Yes. Deploy the runtime on your own CPU servers next to your telephony systems. Inference stays on your local network and audio does not leave your data centre.

Can it run on a phone?

Yes. In Detectif.ai’s deployment study, a detector running CPU-only on a OnePlus handset processed 3,986 of 4,000 clips, 99.65%, faster than real time. That is a measured configuration, not a guarantee for every device.

How much does CPU deepfake detection cost?

On the hosted platform, deepfake detection is $0.0002 per second of audio analysed, and new accounts get 2,000 free credits. For your own CPU servers or a private deployment, talk to the Detectif.ai team.

Run deepfake detection on the CPUs you already have.

Start on the hosted platform, or talk to us about your own servers.