LetterLiGo/Inaudible-Adversarial-Perturbation-Vrifle

[NDSS'24] Inaudible Adversarial Perturbation: Manipulating the Recognition of User Speech in Real Time

19
/ 100
Experimental

This project helps security researchers and system testers evaluate the robustness of Automatic Speech Recognition (ASR) systems like Deep Speech 2. It takes a raw audio file and a target sentence, then generates an 'inaudible' audio perturbation that, when added to the original, makes the ASR system transcribe the target sentence instead of the original spoken content. An adversarial machine learning researcher or a red team security professional would use this to probe vulnerabilities in voice-controlled systems.

No commits in the last 6 months.

Use this if you need to create 'inaudible' audio attacks to manipulate what an Automatic Speech Recognition (ASR) system transcribes from spoken input.

Not ideal if you are looking for a general-purpose audio manipulation tool or an ASR system itself, as this is specifically for adversarial attacks.

speech-recognition-security adversarial-audio voice-system-testing acoustic-cybersecurity red-teaming
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 8 / 25
Maturity 8 / 25
Community 3 / 25

How are scores calculated?

Stars

56

Forks

1

Language

Python

License

Last pushed

Sep 28, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/voice-ai/LetterLiGo/Inaudible-Adversarial-Perturbation-Vrifle"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.