Web Analytics

VeriVoice

The world's first explainable AI speech detector

VeriVoice separates authentic speech from synthesized or cloned voice in seconds. Upload any voice clip and receive a clear authenticity verdict backed by calibrated probability scores, natural-language reasoning, and evidence areas in the spectrogram for faster review.

VeriVoice pairs detection with natural-language reasoning and evidence areas in the spectrogram for synthesized and cloned speech review.

AI voice analysis visual
Performance & Coverage

Built for High-Confidence AI Voice Detection

As the world's first explainable AI speech detector, VeriVoice combines high-accuracy synthesized/cloned voice detection with transparent, reviewable evidence so teams can make faster and more defensible decisions.

96.2%

Balanced Accuracy

Explainable Detector

For AI Voice

Audit-Ready

Evidence & Reports

Model Coverage

VeriVoice is designed to detect outputs from mainstream speech synthesis and cloning pipelines, including modern TTS, voice cloning, voice conversion, and neural codec-based systems. Coverage is continuously updated to keep pace with new model releases, voices, languages, and generation styles.

Real Example

Explainability Analysis Snapshot

A real 4-second slice used in VeriVoice explainability with an AI probability of 100%.

Full Clip + Explainability Slice

Listen to the full clip first, then review the selected 4-second slice used for explainability.

Full Clip

0:00 / 0:00

Selected Slice (3.76s - 7.76s)

0:00 / 0:00

Spectrogram + Key Evidence Areas

Highlighted regions show the cues the detector relies on to make its decision.

VeriVoice explainability spectrogram with highlighted evidence areas

Explainability Analysis

Decision Note: Final AI/Human decision is based on the mean prediction across multiple random 4-second slices. The detailed analysis below is for one selected slice starting from 3.76s to 7.76s.

This audio slice is classified as AI-generated speech based on multiple acoustic artifacts inconsistent with natural human vocal patterns.

Key Evidence Areas

  • An over-smoothed formant transition in the mid-frequencies (~1.0-3.0 kHz) at t=1.25s in-slice (abs=5.01s). This creates a synthetic quality, as human speech typically has more subtle and complex variations when moving between vowel sounds. [weak]
  • A robotic high-frequency texture appearing around t=2.26s in-slice (abs=6.02s), most prominently above 7.0 kHz. This unnatural, static-like energy is a common byproduct of some synthesis methods and lacks the organic complexity of human sibilance or breath sounds. [strong]
  • An abrupt energy jump in the low frequencies (~0-1.0 kHz) at t=3.01s in-slice (abs=6.77s). This sudden, sharp increase in volume at the start of a word is atypical, as human speakers naturally have a smoother, more gradual onset of voicing. [moderate]

The combined presence of these artifacts across different frequency ranges provides high confidence in the AI-generated classification.

Built for Review Teams

Detect Synthetic Voice with Clear, Actionable Evidence

Every VeriVoice result includes an AI/human verdict, calibrated probability, natural-language reasoning, spectrogram evidence regions, and exportable records your team can review, share, and act on.

Verdict + probability Explainability evidence Audit-ready exports
Step 1

Upload and Analyze

Upload WAV, MP3, or FLAC voice clips. VeriVoice normalizes audio and runs detector inference optimized for synthesized and cloned speech artifacts.

Step 2

Review Probability and Reasoning

Receive a clear AI/human verdict, confidence score, and natural-language explanation for faster triage and more consistent analyst decisions.

Step 3

Take Action with Evidence

Export JSON/CSV outputs, attach evidence to review workflows, and preserve a reliable trail for disputes, partner escalations, and compliance.

Integrations

Plug VeriVoice into Your Existing Workflow

Run VeriVoice in the browser, via API, or embedded in internal moderation and trust workflows with webhook callbacks. Enterprise options support larger batches and dedicated review operations.