Real-Time Synthetic Voice Detection Comes to Giga

Today we are launching real-time synthetic voice detection on the Giga platform. Every call handled by a Giga voice agent can now be screened, as the conversation happens, for signs that the speaker on the other end is not a person but a machine.

Why this matters now

Voice used to be the strongest trust signal a business had. If someone called in and sounded like a customer, they almost certainly were one. That assumption has quietly stopped being true, because modern text-to-speech systems can clone a convincing voice from just a few seconds of reference audio¹, and they are cheap, fast, and available to anyone.

The consequences are no longer hypothetical. Criminals used a cloned executive voice to authorize a fraudulent transfer of $243,000 as far back as 2019², and in 2024 a finance employee in Hong Kong wired roughly $25 million after a video call staffed entirely by deepfaked colleagues³. Deloitte’s Center for Financial Services projects that generative AI could push fraud losses in the United States to $40 billion by 2027⁴.

Contact centers face a newer, quieter version of the same attack. Instead of one dramatic heist, bad actors can point AI voice agents at customer support lines and run them as persistent callers: placing many routine calls, probing the refund policy, accepting escalating politely, and retrying across interactions until a rare policy edge case pays out. Any single call looks routine, but taken together this is industrialized fraud running at a scale no human caller could sustain. We built this product because our customers, including some of the largest consumer brands in the world, asked us to help them stop it.

Why synthetic voice detection is hard

The obvious approach sounds simple enough: train a classifier that separates real speech from generated speech. The research community has been working on exactly this problem for a decade through the ASVspoof challenge series⁵, and its results explain why the problem resists easy solutions.

The first obstacle is generalization. Detectors that score near-perfectly against the synthesizers they were trained on degrade sharply when they meet voices from new, unseen models. Müller et al. found that error rates of published detectors increased severalfold when evaluated on in-the-wild audio instead of curated benchmarks⁶. New voice models ship every month, so a detector that memorizes the artifacts of today’s synthesizers is obsolete on arrival. It has to learn what makes speech human rather than what makes one particular model sound fake.

The phone network itself works against detection. Much of the research literature is built on clean, high-sample-rate audio, while real calls are narrowband, compressed by lossy codecs, and degraded by packet loss and background noise. The ASVspoof 2021 evaluation showed that these transmission effects can substantially weaken common detection systems⁷, since many of the spectral artifacts that give away synthetic speech in the lab live in exactly the frequencies the phone network throws away.

Timing adds a further constraint. A forensic model that analyzes a complete recording after the fact is useful for audits, but it cannot stop a live fraud attempt. In a contact center, detection has to run on streaming audio and produce a confident signal within the first moments of a conversation, while the caller is still mid-sentence, without adding any latency to the call.

Watermarking, finally, covers only part of the threat. Some modern voice models embed watermarks in their output, and proactive schemes such as localized audio watermarking are a promising research direction⁸. A watermark, however, only helps when the vendor that generated the audio cooperates. Open-source models, fine-tuned clones, and audio that has been re-recorded or re-encoded all fall outside that protection, so a real defense has to catch synthetic speech that was never watermarked at all.

What we are launching

Giga’s synthetic voice detection runs natively inside our voice platform, with no recording to upload and no separate pipeline to build.

During a live call, the system continuously analyzes the caller’s audio stream and produces a calibrated synthetic-speech signal in real time. It is built for the conditions contact centers actually operate in: narrowband telephony audio, noisy lines, and callers who speak for only a few seconds at a time. Because it evaluates properties of speech production rather than the fingerprint of any single synthesizer, it is designed to hold up against voice models it has never seen.

What happens with that signal is up to each team. A flagged call can be routed to step-up verification, handed to a human specialist, restricted from sensitive actions like refunds or account changes, or simply logged for fraud analytics. Detection results flow into the same infrastructure that already power Giga voice agents, so fraud and CX teams see the fraud detection results.

Our approach draws on the same lines of research that have defined the field, including end-to-end raw-waveform models⁹ and graph attention architectures¹⁰ alongside self-supervised speech representations¹¹, which have consistently set the strongest results in the ASVspoof evaluations. We combine these techniques with training pipelines built specifically for telephony conditions and adversarial robustness, and we continuously refresh them as new synthesis models appear.

Real-time synthetic voice detection is available now for Giga customers. If your support line is seeing repeat-caller fraud, policy abuse, or suspected AI callers, reach out to your Giga team or contact us to see it on your own traffic.