Dashboard

ECAPA-TDNN Voice Authentication System

Total Drivers
—
Voice Enrollments
—
Verifications
—
Spoof Attempts
—

Quick Actions

Register Driver

Create a new driver profile before enrolling their voice

Enroll Voice

Record a voice sample to create a biometric voiceprint

Step 1: Select Driver

Verify Speaker

Verify a speaker's identity against their enrolled voiceprint

Step 1: Select Driver

All Drivers

Manage registered drivers and their voice enrollments

Name Email Status Enrollments Last Enrolled Actions
Loading drivers...

Settings

Configure ECAPA-TDNN parameters and system thresholds

Loading configuration...

Database Schema

The system uses a relational schema with the following tables:

drivers

  • id UUID PK
  • first_name, last_name
  • email UNIQUE
  • phone, driver_license
  • status (active / suspended / inactive / pending)
  • metadata JSONB

voice_enrollments

  • id UUID PK
  • driver_id FK → drivers
  • embedding DOUBLE PRECISION[]
  • phrase_text, audio_hash
  • quality_score, is_active
  • audio_features JSONB

verification_attempts

  • id UUID PK
  • driver_id, enrollment_id FKs
  • similarity_score, threshold_used
  • result (match / no_match / spoof / error)
  • liveness_passed, replay_detected
  • audio_features JSONB

anti_spoof_logs

  • id UUID PK
  • verification_id, driver_id FKs
  • spoof_type (replay / synthetic / channel / ...)
  • confidence
  • details JSONB

Anti-Spoof Controls

Fine-tune detection parameters to prevent voice spoofing

Loading anti-spoof configuration...

Parameter Guide

Speaker Matching

Controls the cosine similarity threshold for determining if two voice embeddings belong to the same speaker. Higher values require a closer match but may increase false rejections.

Audio Quality Gates

Minimum audio duration, RMS energy, and crest factor checks ensure the submitted audio contains genuine speech. These reject silence, extremely short clips, and heavily compressed or synthetic audio.

Anti-Spoofing

Spectral flatness detects synthetic or white-noise audio. Replay detection uses SHA-256 hashing to prevent reuse of previously submitted recordings. Phrase matching verifies the speaker read the correct challenge text.

Enrollment Limits

Controls the maximum number of voice enrollments per driver. Multiple enrollments improve recognition accuracy by capturing voice variation across sessions.

User Management

Manage application users and roles

Users

Username Display Name Role Status Last Login Actions

System Logs

Real-time structured event log

Auto-refresh
Timestamp Type Driver Duration Result Details
Loading logs...
Liveness Check

Say the color and number out loud:

—
Listening...
Time remaining: 3s