Precision-Varying Prediction (PVP): Robustifying ASR systems against adversarial attacks

Matías Pizarro, Raghavan Narasimhan, Asja Fischer
3/23/2026
cs.LGcs.CReess.AS

No official code found. Best implementation: blindconf/PVP_Precision_varying_prediction (0 stars, MIT, last commit 2026-03-23).

Abstract

With the increasing deployment of automated and agentic systems, ensuring the adversarial robustness of automatic speech recognition (ASR) models has become critical. We observe that changing the precision of an ASR model during inference reduces the likelihood of adversarial attacks succeeding. We take advantage of this fact to make the models more robust by simple random sampling of the precision during prediction. Moreover, the insight can be turned into an adversarial example detection strategy by comparing outputs resulting from different precisions and leveraging a simple Gaussian classifier. An experimental analysis demonstrates a significant increase in robustness and competitive detection performance for various ASR models and attack types.

View on arXivView PDF

Code Implementations(1)

blindconf/PVP_Precision_varying_predictionCommunity74%
MIT

[Paper Under Revision] Precision-Varying Prediction (PVP): Robustifying ASR systems against adversarial attacks.

00Feb 27, 20267 months agoMIT
adversarial-attack-and-defenseadversarial-robustnessautomatic-speech-recognition

Similar Papers

MIXPGD: Hybrid Adversarial Training for Speech Recognition Systems

89% match2023

Rethinking Textual Adversarial Defense for Pre-trained Language Models

88% match2022

Evaluation of Adversarial Robustness in Arabic Language Models

86% match2026

Defending against Adversarial Audio via Diffusion Model

86% match1 repo352023

Confidence-Calibrated Adversarial Training: Generalizing to Unseen Attacks

86% match1 repo102019

The Best Defense is Attack: Repairing Semantics in Textual Adversarial Examples

86% match2023

Cite this paper

@article{pizarro2026precisionvarying,
  title  = {Precision-Varying Prediction (PVP): Robustifying ASR systems against adversarial attacks},
  author = {Matías Pizarro and Raghavan Narasimhan and Asja Fischer},
  year   = {2026},
  eprint = {2603.22590},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2603.22590v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2603.22590.svg)](https://codeofpaper.com/paper/2603.22590)

Discussion