Detect synthetic voice
in contact centers.
Protect contact centers against voice cloning and AI deepfakes. VIONIS evaluates acoustic vocoder artifacts, conversational turn latency, and passive semantic probes in real-time call streams.
More than 3 detection layers.8 kHz telephony audioDual-channel separationAcoustic, behavioral & probesStandardized JSON risk score
Voice cloning bypasses
traditional authentication.
Generative voice models clone customer voices with seconds of reference audio. Robust verification requires analyzing physical acoustic anomalies, live turn-taking dynamics, and contextual reasoning.
Generative voice cloning
Synthetic AI models replicate vocal timbre and cadence. Perceptual familiarity alone is insufficient to authenticate high-risk requests.
Telephony compression artifacts
Band-limited 8 kHz telephony codecs (G.711) mask synthetic anomalies, requiring isolated dual-channel analysis.
Conversational latency anomalies
Automated voice bots exhibit anomalous turn delays, rigid response timing, and scripted semantic failures under interruption.
More than 3 complementary
verification layers.
Vionis evaluates acoustic vocoder artifacts, conversational dynamics, and passive semantic probes across Flash, Turbo, and Max.
Spectral signature & neural vocoder analysis
Evaluates harmonic distribution, vocoder spectral flatness, and fine-tuned wav2vec2 XLS-R neural embeddings to detect synthetic voice synthesis and cloning.
Four-stage telephony
verification pipeline.
Submit stereo WAV call recordings and receive an authoritative caller classification from Vionis models (Flash, Turbo, or Max).
Analyze call audio
for synthetic indicators.
Upload a dual-channel WAV recording and let Vionis models (Turbo, Max, or Flash) classify Caller CH0 using acoustic, conversational, and neural evidence.
Live Vionis detection suiteCall inspector
Analyze Caller CH0 with Agent CH1 as conversation context.
Ready when you are
Choose a call recording and model, then run analysis.
Model classification does not verify caller identity. Confidence is the score for the returned verdict.
Caller CH0 is classified. Agent CH1 provides turn-taking context.
Acoustic features and conversational timing, with a degraded-line fallback.
Run analysis submits the original WAV to Flash. Results stay in this session.
REST API for contact center
fraud prevention.
Integrate synthetic voice detection into your SIP gateway, IVR, or agent desktop. Submit call audio to /detect (Flash), /turbo/detect (Turbo), or /max/detect (Max) for a verdict, or /{model}/explain for measured features and model diagnostics.
/turbo/detectLIVE VIONIS API{ "call_id": "call-001", "audio_base64": "<complete stereo WAV as base64>", "sample_rate": 8000, "channels": 2}Verify caller authenticity in real time.
Test sample calls in the console, inspect layer evidence, or integrate directly via the REST API.
Launch dashboard