Secure autonomous systems

Secure Voice and Event Intelligence for AVs

How can an autonomous vehicle verify who is speaking and preserve trustworthy evidence when an incident occurs?

Published in Electronics, 2023

Overview

Can voice identity, spoofing resistance, and distributed accident evidence be designed as one autonomous-vehicle trust architecture?

Voice control creates an intuitive interface for autonomous vehicles, but it also creates identity and spoofing risks. After an accident, accountability introduces a second challenge: event records must remain transparent, verifiable, and resistant to tampering.

This project connects robust speaker identification, ensemble speaker verification, blockchain-based event recording, and location-aware witness voting in one security and accountability framework.

Complete secure autonomous vehicle speaker identification and blockchain event-recording workflow

The complete system workflow links authorized-speaker verification, vehicle control, accident detection, witness voting, and blockchain-backed event recording.

Methods

Authorized speaker identification

Spectral voice features and classification models restrict control to enrolled users instead of accepting commands from any speaker.

Anti-spoofing ensemble

Multiple speaker-verification decisions are combined to improve resilience against synthesized or converted voice attacks.

Blockchain event recording

Smart contracts store vehicle, speaker, time, location, and witness-vote evidence as transparent, tamper-resistant accident records.

3integrated trust mechanisms

4blockchain platforms compared

Smart contractsfor event and witness records

Location awarevehicle witness voting

Contributions

Speaker identification for limiting autonomous-vehicle control to authorized users.
An ensemble verification strategy for detecting voice spoofing attempts.
A blockchain workflow for immutable accident and speaker-event records.
A concept in which nearby vehicles act as independent witnesses and vote on liability.
Latency and simulation analysis for the proposed recording and adjudication workflow.

Speaker identification system model for autonomous vehicle access control — The speaker-identification model converts a driver utterance into acoustic features and an identity decision before permitting access to vehicle functions.

Interpretation

Security before an incident and accountability after it are usually treated separately. This system connects them through identity: the vehicle verifies who issued a command, preserves the associated event evidence, and incorporates observations from nearby vehicles when liability must be investigated.

Current scope

The paper establishes and simulates the integrated architecture. Deployment on production vehicles would require manufacturer interoperability, dependable vehicle-to-network communication, sensor validation, and shared standards for accident evidence.

Additional figures

Cellular vehicular network connecting autonomous vehicles and infrastructure — The cellular vehicular network provides the communication layer through which accident vehicles, witness vehicles, roadside infrastructure, and remote services exchange evidence.

Blockchain records for accident voting and speaker identification — Separate blockchain records preserve witness votes and speaker-identification events, creating an auditable history for safety investigations.

Voice spoofing attack against an autonomous vehicle — The threat model shows how a replayed or synthesized voice can impersonate an authorized driver, motivating the project's identity-verification layer.

Authorized speaker identification concept for autonomous vehicles — Speaker identification narrows control privileges to enrolled users so that a spoken command is evaluated for both content and identity.

Levels of vehicle automation from driver assistance to full automation — The automation taxonomy situates the security problem across the transition from human-controlled vehicles to fully autonomous operation.

Comparison of blockchain platforms for autonomous vehicle event recording — The platform comparison examines the characteristics of candidate blockchains for transparent and tamper-resistant vehicle records.

Simulation of accident vehicles, witness vehicles, and normal traffic — The simulation distinguishes accident participants, nearby witnesses, and normal vehicles while varying their positions and speeds.

Blockchain event-recording latency under changing vehicle and witness counts — Latency experiments show how the accident-recording workflow changes as the number of participating vehicles and witnesses grows.