WaveDetect: Robust Framework for Machine-Generated Text Detection via Wavelet Transform

2026-06-22

AI Summary

Existing text detection methods suffer from insufficient robustness under adversarial perturbations, cross-domain shifts, and the rapid evolution of large language models (LLMs). This work proposes the first integration of differentiable continuous wavelet transform into text detection, converting model output probability sequences into time-frequency spectrogram representations. This approach uncovers a stable, human-imperceptible “spectral fingerprint” inherent in machine-generated text. By leveraging time-frequency analysis to capture intrinsic patterns of synthetic content, the method achieves new state-of-the-art performance across multiple benchmarks—including RAID, EvoBench, and Domain-Shift—significantly improving detection accuracy and generalization against adversarial attacks, out-of-distribution topics, and text generated by emerging LLMs.

Abstract

As Large Language Models asymptotically approach human-level fluency in natural language generation, solely relying on surface-level semantic artifacts for detecting LLM-generated texts has become increasingly precarious. Existing detectors often falter when facing three critical challenges: adversarial perturbations, cross-domain shifts, and the rapid temporal evolution of the foundation model. To address these issues, we propose \wavedetect, a novel framework that reformulates text detection as a signal processing task within the time-frequency domain. Unlike previous methods that analyze static token probability distributions, \wavedetect models the generated output as a probability signal, upon which a differentiable Continuous Wavelet Transform is applied to convert them into learnable spectral representations. This process reveals the intrinsic ``spectral fingerprints'' in machine-generated texts--patterns that remain invisible in time domain. Comprehensive evaluations on three well-curated datasets (RAID, EvoBench, and Domain-Shift) show that our method achieves a new state-of-the-art. It not only achieves superior accuracy but also exhibits remarkable robustness against sophisticated attacks, generalization across out-of-distribution topics and unseen evolving LLMs. Our results validate the efficacy of spectral analysis as a promising paradigm for LLM-generated texts detection.

Problem

• machine-generated text detection
• adversarial perturbations
• cross-domain shifts
• temporal evolution of LLMs

Innovation

• Wavelet Transform
• Machine-Generated Text Detection
• Spectral Fingerprint
• Time-Frequency Domain
• Robustness